Mixed platoon control of automated and human-driven vehicles at a signalized intersection: dynamical analysis and optimal control
Extracted source text ยท arXiv 2010.16105
\documentclass[preprint,10pt,3p]{elsarticle}
%preprint,10pt,3p
\usepackage{lineno,hyperref}
\usepackage{graphics} % for pdf, bitmapped graphics files
\usepackage{epsfig} % for postscript graphics files
%\usepackage{mathptmx} % assumes new font selection scheme installed
\usepackage{times} % assumes new font selection scheme installed
\usepackage{amsmath} % assumes amsmath package installed
\usepackage{amssymb} % assumes amsmath package installed
\usepackage{amsfonts}
%\usepackage{cite}
\usepackage{epstopdf}
\usepackage{subfigure,color,balance}
\usepackage{verbatim}
\usepackage{cases}
\usepackage{enumerate}
%\usepackage{enumitem}
\usepackage{balance}
\usepackage{framed}
\usepackage{booktabs}
\usepackage{color}
\usepackage{cancel}
\usepackage{epstopdf}
\usepackage{booktabs}
\usepackage{multirow}
\usepackage{makecell}
\usepackage{bm}
\usepackage{diagbox}
\graphicspath{{figures/}}
\newcommand{\blue}[1]{{\color{blue}#1}}
\newcommand{\red}[1]{{\color{red}#1}}
\newcommand{\etal}{\textit{et al}.}
\newcommand{\ie}{\textit{i}.\textit{e}.}
\newcommand{\eg}{\textit{e}.\textit{g}.}
\newtheorem{definition}{Definition}
\newdefinition{remark}{Remark}
\newtheorem{lemma}{Lemma}
\newtheorem{theorem}{Theorem}
%\newtheorem{assumption}{Assumption}
%\newtheorem{proposition}{Proposition}
%\newtheorem{corollary}{Corollary}
%\newtheorem{problem}{Problem}
\newtheorem{step}{Step}
\newproof{proof}{Proof}
%\modulolinenumbers[5]
\journal{Transportation Research Part C: Emerging Technologies}
%%%%%%%%%%%%%%%%%%%%%%%
%% Elsevier bibliography styles
%%%%%%%%%%%%%%%%%%%%%%%
%% To change the style, put a % in front of the second line of the current style and
%% remove the % from the second line of the style you would like to use.
%%%%%%%%%%%%%%%%%%%%%%%
%% Numbered
%\bibliographystyle{model1-num-names}
%% Numbered without titles
%\bibliographystyle{model1a-num-names}
%% Harvard
%\bibliographystyle{model2-names.bst}\biboptions{authoryear}
%% Vancouver numbered
%\usepackage{numcompress}\bibliographystyle{model3-num-names}
%% Vancouver name/year
%\usepackage{numcompress}\bibliographystyle{model4-names}\biboptions{authoryear}
%% APA style
\bibliographystyle{model5-names}\biboptions{authoryear}
%% AMA style
%\usepackage{numcompress}\bibliographystyle{model6-num-names}
%% `Elsevier LaTeX' style
%\bibliographystyle{elsarticle-num}\biboptions{compress}
%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\begin{frontmatter}
\title{Mixed platoon control of automated and human-driven vehicles at a signalized intersection: dynamical analysis and optimal control}
% Signalized intersection control via mixed platoon with automated and human-driven vehicles: dynamical analysis and optimal control
%\title{Elsevier \LaTeX\ template\tnoteref{mytitlenote}}
%\tnotetext[mytitlenote]{Fully documented templates are available in the elsarticle package on \href{http://www.ctan.org/tex-archive/macros/latex/contrib/elsarticle}{CTAN}.}
%% Group authors per affiliation:
%\author{Chaoyi Chen, Qing Xu, Jiawei Wang, Jianqiang Wang and Keqiang Li}
%%\fnref{myfootnote}
%\address{School of Vehicle and Mobility, Tsinghua University, Beijing 100084, China}
%%\fntext[myfootnote]{Since 1880.}
%% or include affiliations in footnotes:
\author[a,b]{Chaoyi Chen}
%\ead{chency19@mails.tsinghua.edu.cn}
\author[a,b]{Jiawei Wang}
%\ead{wang-jw18@mails.tsinghua.edu.cn}
\author[a,b]{Qing Xu}
%\ead{qingxu@tsinghua.edu.cn}
\author[a,b]{Jianqiang Wang}
%\ead{wjqlws@tsinghua.edu.cn}
\author[a,b]{Keqiang Li\corref{cor1}}
\ead{likq@tsinghua.edu.cn}
\cortext[cor1]{Corresponding author: Keqiang Li}
%\fnref{fn1}
%\fntext[fn1]{This is the first author footnote.}
\address[a]{School of Vehicle and Mobility, Tsinghua University, Beijing, 100084, China}
\address[b]{Tsinghua University-Didi Joint Research Center for Future Mobility, Beijing, 100084, China}
%, which shows that the mixed platoon is stable and controllable under a mild condition regardless of the platoon size $ n $
%The terminal velocity is further optimized based on the proposed mixed platoon formation.
\begin{abstract}
The emergence of Connected and Automated Vehicles (CAVs) promises better traffic mobility for future transportation systems. Existing research mostly focused on fully-autonomous scenarios, while the potential of CAV control at a mixed traffic intersection where human-driven vehicles (HDVs) also exist has been less explored. This paper proposes a notion of ``$ 1+n $" mixed platoon, consisting of one leading CAV and $ n $ following HDVs, and formulates a platoon-based optimal control framework for CAV control at a signalized intersection. Based on the linearized dynamics model of the ``$ 1+n $" mixed platoon, fundamental properties including stability and controllability are under rigorous theoretical analysis. Then, a constrained optimal control framework is established, aiming at improving the global traffic efficiency and fuel consumption at the intersection via direct control of the CAV. A hierarchical event-triggered algorithm is also designed for practical implementation of the optimal control method between adjacent mixed platoons when approaching the intersection. Extensive numerical simulations at multiple traffic volumes and market penetration rates validate the greater benefits of the mixed platoon based method, compared with traditional trajectory optimization methods for one single CAV.
\end{abstract}
\begin{keyword}
Connected and Automated Vehicle \sep Signalized intersection \sep Mixed traffic system \sep Optimal control
%\texttt{elsarticle.cls}\sep \LaTeX\sep Elsevier \sep template
%\MSC[2010] 00-01\sep 99-00
\end{keyword}
\end{frontmatter}
%\linenumbers
%Traffic congestion has been a growing concern %caused by imperfect driving behavior and traffic guidance is commonly seen
%in most urban areas around the world, causing great loss of travel efficiency and fuel consumption. In
\section{Introduction}
%\subsection{Motivation}
%{\color{red}
%According to the National Highway Traffic Safety Administration (NHTSA), road crashes caused 36,560 fatalities and 2,710,000 injuries in 2018 in the United States~\cite{NHTSA}.
%According to U.S. Department of Transportation (DOT), an average of one-quarter of traffic fatalities and roughly half of all traffic injuries are attributed to intersections~\cite{DOT}. %This indicates that
As planned points of conflict in urban traffic networks, intersections play a critical role in traffic mobility optimization. % What's more, urban traffic congestion resulted in a total economic loss of 166 billion dollars in 2017 in US~\cite{schrank2019}.
%To address the problem, one of the typical methods is to optimize traffic signal phase and timing (SPAT) at intersections~\cite{zhao2011computational}. % and the effectiveness of various techniques has been investigated for decades.
Existing research has shown that the frequent stop-and-go and idling behavior of individual vehicles when approaching the intersection is the main cause of traffic congestion and casualties~\cite{lee2012development, rakha2001estimating}. Accordingly, trajectory optimization for individual vehicles at the intersection has attracted significant attention.
%For example, Beijing citizen get stuck in traffic jam for 2 hours 50 min per day on average in 2018~\cite{Beijing_Traffic_2019}.}
%With the recent development of wireless communication and autonomous control technologies, the emergence of Connected and Automated Vehicles (CAVs) is expected to provide new opportunities to improve traffic mobility at intersections. %Previous researcher finds that CAV technology can reduce 30\% of the traffic accidents, improve traffic efficiency by 10\%, and cut off 5\% of the fuel consumption and emission~\cite{Li17}. %including optimal control~\cite{Schutter98}, fuzzy neural network~\cite{Ella01}, and reinforcement learning~\cite{Duan10}.
In particular, the emergence of Connected and Automated Vehicles (CAVs) has provided new opportunities for improving the traffic performance at intersections. Compared to traditional human-driven vehicles (HDVs), CAVs can acquire accurate information of surrounding traffic participants and traffic signal phase and timing (SPAT) through vehicle-to-vehicle and vehicle-to-infrastructure communication~\cite{contreras2017internet}, and their velocity trajectories in approaching intersections can be directly optimized in the pursuit of higher traffic efficiency and lower fuel consumption. Multiple methods have been applied to CAV control at signalized intersections, including model predictive control~\cite{asadi2011predictive,yang2017eco}, fuzzy logic~\cite{Milanes10,onieva2015multi} and optimal control~\cite{li2015eco,jiang2017eco}. Moreover, the potential of cooperative control of traffic signals and CAVs has also been recently discussed~\cite{xu2018cooperative}.%; see, \eg,~\cite{xu2017v2i}, which proposed a hierarchical structure to decouple the traffic SPAT and CAV's driving strategy.
%Regarding nonsignalized intersections, several distributed optimal control frameworks have also been established for CAVs to minimize the fuel consumption and meanwhile maximize the traffic capacity; see, \eg,~\cite{malikopoulos2018decentralized}.
%{\color{green}replace~\cite{zhang2016optimal, zhang2017decentralized, zhang2017optimal} by journal papers.}
%One of the most important difficulties in nonsignalized intersections is to optimize the passing order of CAVs. Spanning tree~\cite{li2006cooperative, xu2018distributed}, Monte Carlo tree search ~\cite{xu2019cooperative} and dynamic programming ~\cite{yan2009autonomous} are proved to be effective in tackling the problem.
%{\color{red}*Might remove the following statements of data, since in the main text, we assume a fully-connected setup and the statements below is distractive. The paragraph above has presented the characteristics of CAVs.*}
%{\color{blue}[Delete*] For real-world implementation of CAVs, essential requirements include equipment of connected devices, \eg, being equipped with vehicle-to-vehicle or vehicle-to-infrastructure equipment~\cite{contreras2017internet}, and capabilities of autonomous control, \ie, being directly operated by industrial personal computers or micro-controllers~\cite{guanetti2018control}. According to the Department of Transportation of US, it will take 20-25 years to realize 95\% MPR of connected vehicles~\cite{John08}, and based on the estimation of Ministry of Industry and Information Technology of China, the MPR of partial/conditional autonomous vehicles is expected to reach 25\% by 2025~\cite{Auto_industry}. Considering the long-term deployment of CAV technologies, research on mixed traffic environment of CAVs and HDVs is of great importance to improve traffic mobility at intersections in the near future.}
%\subsection{Related Work}
The aforementioned research mainly considered a fully-autonomous scenario---the market penetration rate (MPR) of CAVs is 100\%. In practice, however, it might take decades for all the HDVs in current transportation systems to be transformed into CAVs. Instead, a more practical scenario in the near future is a mixed traffic system where CAVs and HDVs coexist~\cite{zheng2020smoothing}. Existing research on mixed traffic intersections mostly focused on the estimation of traffic states and optimization of traffic signals~\cite{Priemer09, Salman17, zheng2017estimating, feng2015real}. For example, Priemer \etal~utilized dynamic programming to estimate the queuing length in the mixed traffic environment and then optimized the traffic phase time based on the estimated results~\cite{Priemer09}. Some other methods, including fuzzy logic~\cite{Salman17}, expectation maximization~\cite{zheng2017estimating} and dynamic programming~\cite{feng2018spatiotemporal}, have also been exploited to achieve a similar goal. To improve the estimation accuracy, the celebrated results on microscopic car-following models have been recently employed to describe the behaviors of HDVs, including Gipps Model~\cite{gipps1981behavioural}, Optimal Velocity Model (OVM)~\cite{helbing1998generalized} and Intelligent Driver Model (IDM)~\cite{treiber2013traffic}; see, \eg,~\cite{du2017coordination,fang2020trajectory}.
Despite these existing works, the topic of CAV control, \ie, trajectory optimization of CAVs, in mixed traffic intersections has not been fully discussed. To tackle this problem, several works regarded HDVs as disturbances in the control of CAVs~\cite{du2017coordination,ala2016modeling}, or focused on the task of collision avoidance based on the prediction of HDVs' behaviors~\cite{jiang2017eco,ma2017parsimonious}. It is worth noting that most of these methods were limited to improving the performance of CAVs themselves in their optimization frameworks, instead of optimizing the global traffic flow consisting of both HDVs and CAVs at the intersection. Two notable exceptions are in~\cite{zhao2018platoon,liang2019joint}, which attempted to improve the performance at the signalized intersection from the perspective of the so-called \emph{mixed platoon}. They enumerated several possible formations consisting of HDVs and CAVs, and investigated their effectiveness through small-scale simulation experiments. However, a general and explicit definition of the mixed platoon has not been clarified, and fundamental properties of the mixed platoon at intersections have been less explored. Moreover, a specific optimal control framework for the mixed platoon at intersections with global consideration of improving the entire traffic performance is still lacking.
{In fact, considering the interaction between adjacent vehicles on the same lane, it is easy to understand that velocity trajectories of CAVs could have a certain influence on those of surrounding vehicles, especially the vehicles following behind them. Accordingly, the driving strategies of CAVs could have a direct impact on the performance of the entire mixed traffic intersection; such impact might even be negative when inappropriate CAV strategies are adopted~\cite{ala2016modeling}. By contrast, when taking the performance of entire mixed traffic into explicit consideration, the optimization of CAVs' trajectories could bring further benefits to traffic mobility. Such idea of improving the global traffic performance via controlling CAVs has been recently proposed as \emph{Lagrangian Control} of traffic flow~\cite{stern2018dissipation}, which has been discussed in various traffic scenarios, including closed ring road~\cite{zheng2020smoothing}, open straight road~\cite{wang2020leading}, traffic bottleneck~\cite{vinitsky2018lagrangian}, and non-signalized intersection~\cite{wu2017emergent}. Regarding signalized intersections, the potential of this notion has not been well understood.}
% Therefore, it is possible to improve the performance of the entire mixed traffic intersection via direct control of the CAVs. However, this influence has not been explicitly discussed in the literature on intersections. Among the mixed traffic researches, one notable exception is
% and made further studied on corresponding optimization of the traffic signals
%\subsection{Contributions}
In this paper, we focus on the scenario of a signalized intersection where HDVs and CAVs coexist and aim at improving the performance of the entire mixed traffic intersection through direct control of CAVs. To address this problem, we propose a novel framework which separates the traffic flow into ``$1+n$" microstructures consisting of one leading CAV and $n$ following HDVs. Such microstructure is particularly common in the near future when the MPR is relatively low, which is named as ``$1+n$" mixed platoon. We discuss the possibility of letting the first CAV lead the motion of following $ n $ HDVs in approaching intersections, and show how to enable the CAV to benefit the global traffic mobility in the proposed structure of mixed platoon. First, we present the dynamics model of ``$1+n$" mixed platoon systems based on linearized car-following models and investigate its fundamental properties, including open-loop stability and controllability. Then, we establish the optimal control framework for the mixed platoon and design a hierarchical algorithm to improve the global traffic mobility performance at a signalized intersection. Specifically, our contributions are as follows:
\begin{enumerate}[(1)]
\item The notion of the ``$1+n$" mixed platoon is proposed for trajectory optimization of CAVs at signalized intersections in the mixed traffic environment. Rather than enumerating the mixed platoon formations~\cite{zhao2018platoon,liang2019joint}, we provide an explicit definition of the mixed platoon. Based on the linearized dynamics model, we perform a rigorous theoretical analysis of its fundamental properties, including open-loop stability and controllability. Our theoretical results reveal that the ``$1+n$" mixed platoon is always controllable under a very mild condition, regardless of the platoon size $n$.
\item An optimal control framework is established for the ``$1+n $'' mixed platoon at a constant traffic SPAT intersection scenario. Instead of being limited to optimizing the performance of the CAVs only~\cite{asadi2011predictive,yang2017eco,Milanes10,onieva2015multi,li2015eco,jiang2017eco,du2017coordination,fang2020trajectory,ala2016modeling,ma2017parsimonious}, our optimal control formulation aims at improving the global performance of the entire mixed traffic intersection. Precisely, the velocity deviations and fuel consumption of all the vehicles in the mixed platoon are under explicit consideration. Moreover, unlike~\cite{zhao2018platoon,liang2019joint}, we optimize the terminal velocity setting to maximize the traffic throughput.
\item Finally, a hierarchical algorithm is proposed to accomplish the optimal control framework of ``$1+n $'' mixed platoons, which can be applied in any MPR of mixed traffic environments. An event-triggered mechanism is designed to avoid potential collisions of adjacent mixed platoons. Large-scale traffic simulations are conducted at multiple traffic volumes and MPRs, and it is observed that the proposed mixed platoon based control method surpasses the traditional intersection control method for single CAV~\cite{asadi2011predictive,yang2017eco} in both traffic efficiency and fuel consumption.
\end{enumerate}
%with explicit consideration of HDVs in the optimization of CAVs' . From above we can conclude that in mixed traffic environment research area there is few research focused on how HDVs' behavior influences CAVs' decision, \ie, most of the research didn't consider HDVs explicitly in the optimization, which may have adverse impact in the whole traffic efficiency.
%Our second focus is to A scenario is set as the traffic light operates a constant phase signal, and the mixed traffic flow constantly enters the intersection, as shown in Fig~\ref{fig:IntersectionStructure}. The modeling of "$1+n $" platoon is formulated as an overall optimal control problem. In this case, the CAV collect the information from not only the traffic light and the vehicle in front of it, but also the vehicle following it. The CAV' control input lets the CAV arrive at the stopping line just as the traffic light turn green, meanwhile CAV decelerates in advance to lead the following HDVs pass the intersection in proper velocity.
%\subsection{Organization of This Paper}
The rest of this paper is organized as follows. Section~\ref{Section:ProblemFormulation} introduces the problem of the system modeling for the proposed``$1+n $'' mixed platoon. Section~\ref{Section:Methodology} presents the system dynamics analysis, optimal control framework and algorithm design. The simulation results are shown in Section~\ref{Section:SimulationResult}, and Section~\ref{Section:Conclusion} concludes this paper.
\section{Problem Statement}
\label{Section:ProblemFormulation}
In this section, we firstly introduce the scenario setup, then present the dynamical modeling of individual vehicles and the mixed platoon systems.
\begin{figure}[t]
\centering
\includegraphics[width=0.65\linewidth]{IntersectionStructure.pdf}
\caption{Illustration for the mixed traffic intersection. Red vehicles represent CAVs, which can transmit and receive vehicle information and the vehicle is fully autonomous driving. Black vehicles represent HDVs, which can only transmit ego vehicle information to other vehicles and are controlled by car following model.}
\label{fig:IntersectionStructure}
\end{figure}
\subsection{Scenario Setup}
\label{section:ScenarioSetup}
% Previous works on intersections usually focused on the control of traffic signals phasing time~\cite{Schutter98, Ella01, Wiering00, Abdulhai03, Duan10} or the velocity control of individual vehicles in a fully-autonomous scenario~\cite{malikopoulos2018decentralized,bian2019cooperation,feng2015real,xu2017v2i}.
In this paper, we consider a typical signalized intersection scenario in the mixed traffic environment, where HDVs and CAVs coexist; see Fig.~\ref{fig:IntersectionStructure} for illustration. % Among the four directions at the intersection, {\color{red}each direction has two incoming lanes, where the left lane is only for left-turning vehicles, and the right lane is for both straight-going vehicles and right-turning vehicles}.
A traffic light is deployed in the center to guide individual vehicles to drive through the intersection. Note that CAVs follow the instructions from a \emph{central cloud coordinator}, which collects information from all the involved vehicles around the intersection and calculates the optimal velocity trajectories for each CAV. The design of the control strategies for the central cloud coordinator is presented in Section~\ref{Section:Methodology}.
% zhang2016optimal,zhang2017decentralized,zhang2017optimal
Motivated by previous works on signalized intersections~\cite{feng2015real,malikopoulos2018decentralized,bian2019cooperation}, we separate the intersection into three zones, as illustrated in Fig.~\ref{fig:IntersectionStructure}. The red square area in the center is named as the Merging Zone (MZ), which is the area of potential lateral collision of the involved vehicles. The ring area between two dashed lines is called the Observation Zone (OZ), where the CAVs and HDVs are allowed to perform lane-changing behaviors. The area between OZ and MZ is the Control Zone (CZ), where the CAVs are under direct control from the central cloud coordinator. The specific range of each zone is discussed in Section~\ref{Section:SimulationResult}.
Similar to existing research~\cite{zhao2018platoon,liang2019joint}, the following assumptions are needed to facilitate the control design for signalized intersection controls, as well as the system modeling and dynamics analysis.
\begin{enumerate}[(1)]
\item All the vehicles are connected vehicles, which means that both the CAVs and the HDVs are able to transmit their velocity and position to the central cloud coordinator through wireless communication, \emph{e.g.}, V2I communication~\cite{gerla2014internet}. An ideal communication condition without communication delay or packet loss is under consideration.
\item All the CAVs are capable of fully autonomous driving, which follow the velocity trajectories assigned from the central cloud coordinator after they enter CZ. Regarding the HDVs, they are controlled by human drivers, for which we assume a general car-following model to describe their driving behavior (see Section~\ref{Sec:HDVModel} for details).
\item {Lane changing is not permitted in CZ. As shown in~\cite{ala2016modeling, xu2018distributed}, unexpected lane changing behaviors might worsen the traffic efficiency, especially near the intersection. Therefore, lane changing is only allowed in OZ, while In CZ, we only need to focus on the longitudinal behavior of each vehicle.}
\end{enumerate}
\subsection{Dynamical Modeling of Mixed Platoon Systems}
\label{Sec:HDVModel}
\begin{figure}[!t]
\centering
\includegraphics[width=0.7\linewidth]{InformationTopology.pdf}
\caption{Schematic for the ``$1+n$'' mixed platoon. Red arrows denote the information flow of the leading CAV (colored in red), which collects information from the following HDVs as well as the traffic light and has an external control input. Black arrows represent the information flow of the HDVs (colored in black), which are under control by human drivers and only acquire information from the preceding vehicle, as shown in~\eqref{equ:CFmodel}. {Note that the coordinate origin is set at the stopping line, where the vehicle position is $ x = 0 $.}}
\label{fig:InformationTopology}
\end{figure}
%In this paper, we consider the mixed-traffic environment, where CAVs and HDVs coexist at the intersection.
In this paper, we propose the notion of ``$1+n$'' mixed platoon as shown in Fig.~\ref{fig:InformationTopology}. It consists of one leading CAV and $n$ following HDVs. Specifically, each CAV is designed as the leading vehicle of the mixed platoon, which leads the motion of the following $n$ HDVs with the aim of improving the performance of the entire mixed platoon when passing the intersection.
%CAVs' initiative of improving the traffic efficiency and fuel consumption needs to be exploited. The mixed platoon model we designed is interpreted in %Fig.~\ref{fig:InformationTopology}. Generally speaking, each CAV is designed as the leading vehicle of the mixed platoon, leading the following $n$ %HDVs. %Since each mixed platoon contains 1 CAV and $ n $ HDVs, namely the mixed platoon is referred to as "1+$ n $" platoon. }
Consider the $n$ following HDVs in the mixed platoon illustrated in Fig.~\ref{fig:InformationTopology}. We denote the position and velocity of vehicle $i$ at time $t$ as $x_i(t)$ and $v_i(t)$, respectively. The headway distance of vehicle $i$ from vehicle $i-1$ is defined as ${d}_{i}(t):=x_{i-1}(t)-x_{i}(t) $. Then, $ \dot{v}_{i}(t) $ and $\dot{d}_{i}(t) =v_{i-1}(t)-v_{i}(t)$ represents the acceleration of vehicle $ i $, and the relative velocity of vehicle $i$ from vehicle $i-1$, respectively.
A great many efforts have been made in previous works to describe the HDVs' car-following dynamics, with several significant models developed, \emph{e.g.}, OVM~\cite{helbing1998generalized} and IDM~\cite{treiber2013traffic}. As shown in the literature~\cite{zheng2020smoothing,wilson2011car}, most of them can be expressed as the following general form ($i=1,\ldots,n$)
\begin{equation}
\label{equ:CFmodel}
\dot{v}_{i}(t) =F\left(d_{i}(t), \dot{d}_{i}(t), v_{i}(t)\right),
\end{equation}
which means that the acceleration of vehicle $i$ is determined by its headway distance, relative velocity and its own velocity.
In this paper, we require that the mixed platoon passes the intersection at a pre-specified equilibrium velocity $v^{\ast}$. In equilibrium traffic state, we have $ \dot{d}_{i}(t) = 0 $ for $i=1,\ldots,n$, and thus each vehicle has a corresponding equilibrium headway distance $ d^{\ast} $, where it holds that
\begin{equation}
\label{equ:CFmodel:Stable}
F\left(d^{\ast}, 0, v^{\ast}\right) = 0.
\end{equation}
Then we employ the deviation of the current state $(d_i(t),v_i(t))$ of vehicle $i$ from the equilibrium state $(d^{\ast},v^{\ast})$ as its state variable, given by
\begin{equation}
\begin{cases}
\label{equ:ErrorState}
\tilde{d}_{i}(t) =d_{i}(t)-d^{*}, \\
\tilde{v}_{i}(t) =v_{i}(t)-v^{*}.
\end{cases}
\end{equation}
Applying the first-order Taylor expansion to~\eqref{equ:CFmodel} leads to the linearized dynamics of HDVs around the equilibrium state $(d^{\ast},v^{\ast})$, given as follows ($i=1,\ldots,n$)
\begin{equation}
\label{equ:ErrorSpeedDiff}
\begin{cases}
\dot{\tilde{d}}_i(t)= v_{i-1}(t)-v_i(t),\\
\dot{\tilde{v}}_{i}(t) =\alpha_{1} \tilde{d}_{i}(t)-\alpha_{2} \tilde{v}_{i}(t)+\alpha_{3} \tilde{v}_{i-1} (t),
\end{cases}
\end{equation}
with
$
\label{equ:ErrorSpeedDiff2}
\alpha_{1}=\frac{{\partial} F}{{\partial} d},\,
\alpha_{2}=\frac{{\partial} F}{{\partial} \dot{d}}-\frac{{\partial} F}{{\partial} v},\,
\alpha_{3}=\frac{{\partial} F}{{\partial} \dot{d}},
$
evaluated at the equilibrium state $(d^{\ast},v^{\ast})$. For trade-off between model fidelity and computational tractability, we consider the OVM model~\cite{helbing1998generalized} as the specific car-following model in our optimal control formulation in Section~\ref{Section:OptimalControlMode}. Note that other HDV models, \eg, IDM~\cite{treiber2013traffic}, can also be applied to derive the specific expression for~\eqref{equ:ErrorSpeedDiff}. In OVM, the general expression~\eqref{equ:CFmodel} of HDVs' car-following dynamics becomes
\begin{equation}
\label{equ2:OVM_a}
\dot{v}_{i}=\kappa[V_{\mathrm{des}}({d}_{i})-v_i],% \quad \mathrm{for $i \in [1,\mathrm{n}]$},
\end{equation}
where $ V_{\mathrm{des}} $ is the driver's desired velocity at headway distance $ {d}_{i} $, given by
\begin{equation}
\label{equ2:OVM_vopt}
V_{\mathrm{des}}({d}_{i})=V_1+V_2{\tanh}[C_1({d}_{i}-L_\mathrm{veh})-C_2],
\end{equation}
with $ L_{\mathrm{veh}} $ denoting the vehicle length and the rest of the symbols are constants. In this case, the specific value of the coefficients in~\eqref{equ:ErrorSpeedDiff} can be calculated by
\begin{equation}
\label{equ2:OVM_vopt_alphai}
\alpha_{1} = {\kappa}V_2C_1\left\{1-{\tanh}^2[C_1({d}_{i}-L_\mathrm{veh})-C_2]\right\}, \alpha_{2} = \kappa, \alpha_{3} = 0.
\end{equation}
%In equation~\eqref{equ:CFmodel}, the general expression of car-following model is under consideration.
%\subsection{Mixed Platoon Systems Modeling}
Regarding the leading CAV, indexed as vehicle $0$, its acceleration signal is utilized as the control input $u(t)$. Then, the longitudinal dynamics of the leading vehicle can be expressed as the following second-order form
\begin{equation}
\begin{cases}
\label{equ:CAV_State}
\dot{x}_{0}(t) =v_{0}(t), \\
\dot{v}_{0}(t) =u(t) .
\end{cases}
\end{equation}
It is worth noting the acceleration signal of the leading CAV is also the only external control input of the entire system of the mixed platoon illustrated in Fig.~\ref{fig:InformationTopology}. %, since the acceleration of the following HDVs is only determined by the state of the preceding vehicle and ego vehicle, as shown in~\eqref{equ:ErrorSpeedDiff}.
Lumping the state of both the leading CAV and the following HDVs yields the state vector of the entire mixed platoon system, given as follows
\begin{equation}
\label{equ:Platoon_state}
%\begin{split}
{X}(t)
%&=\left[
%\begin{matrix}
%{x}_0 & {d}_1 & \dots & {d}_n
%\end{matrix}
%\right]^ {\rm T}\\
%&
=\left[
\begin{matrix}
x_0(t) & v_0(t) & \tilde{d}_1(t) & \tilde{v}_1(t) &\dots & \tilde{d}_{n}(t) & \tilde{v}_{n}(t)
\end{matrix}
\right]^ {\rm T}.
%\end{split}
\end{equation}
Based on~\eqref{equ:ErrorSpeedDiff} and~\eqref{equ:CAV_State}, the state-space model of the mixed platoon system is obtained
\begin{equation}
\label{equ:MixedPlatoon}
\dot{{X}}(t)={A} {X}(t)+{B} {u}(t),
\end{equation}
where (${A} \in \mathbb{R}^{(2n+2) \times(2n+2)}, \, {B} \in \mathbb{R}^{(2n+2) \times 1}$)
\begin{equation*}
\label{equ:MatrixA}
{A}=\left[ \begin{array}{cccccc}
{{C}_{1}} & {0} & {\cdots} & {\cdots} & {0} & 0 \\
{{A}_{2}} & {{A}_{1}} & {0} & {\cdots} & {\cdots} & {0} \\
{0} & {{A}_{2}} & {{A}_{1}} & {0} & {\cdots} & {0} \\
{\vdots} & {\ddots} & {\ddots} & {\ddots} & {\ddots} & {\vdots} \\
{0} & {\cdots} & {0} & {{A}_{2}} & {{A}_{1}} & {0} \\
{0} & {\cdots} & {\cdots} & {0} & {{A}_{2}} & {{A}_{1}}
\end{array}\right],\;
{B}=\left[ \begin{array}{c}
B_1\\B_2\\B_2\\ \vdots \\B_2\\B_2
\end{array}\right],
\end{equation*}
with
\begin{equation*}
\label{equ:MatrixC}
{A}_{1}=\left[ \begin{array}{cc}
{0} & {-1} \\
{\alpha_{1}} & {-\alpha_{2}}
\end{array}\right], \,
{A}_{2}=\left[ \begin{array}{cc}
{0} & {1} \\
{0} & {\alpha_{3}}
\end{array}\right],\,
{B}_{1}=\left[ \begin{array}{c}
{0} \\
1
\end{array}\right],\,
{B}_{2}=\left[ \begin{array}{c}
{0} \\
0
\end{array}\right],\,
{C}_{1}=\left[ \begin{array}{cc}
{0} & {1} \\
{0} & {0}
\end{array}\right].
\end{equation*}
%\subsection{Problem Statement}
%\begin{comment}
%Malikopoulos~\cite{malikopoulos2018decentralized} \emph{et al.} did marvelous works on how to build up CAVs' velocity trajectories problem at signalized intersection as an optimal control problem. And Xu~\cite{xu2017v2i} built a hierarchical framework of the coordination of traffic light and CAVs. However, HDVs are not considered in their researches.
%\end{comment}
%{\color{red}Could move to the introduction: Some recent research~\cite{zhao2018platoon,liang2019joint} have investigated the potential of forming CAVs and HDVs into mixed platoons in improving traffic performance at intersections. However, this is still no research on the specific proper mixed platoon forming method and the mathematical analysis on the platoon is not fully discussed}.
% Based on the ``$ 1+n $" mixed platoon model~\eqref{equ:MixedPlatoon}, we first discuss some fundamental properties of the mixed platoon system from a control-theoretic perspective. Specifically, the following two notions are under consideration.
%{\color{red} What could controllability and stability tell us. Refer to the TITS paper.}
%\begin{problem}
% Examine the open-loop stability and controllability.
%\end{problem}
%After giving a mathematical proof of the stability and controllability of the proposed "1+$ n $" platoon.
\begin{remark}
Previous work on control of individual vehicles at signalized intersections mostly focused on the dynamics~\eqref{equ:CAV_State} of CAVs only; see, \eg,~\cite{yang2017eco,li2015eco, jiang2017eco}. Considering the interaction between neighboring vehicles, in this paper we focus on the dynamics~\eqref{equ:MixedPlatoon} of the entire mixed platoon system consisting of both the CAV and its following HDVs, and seek to improve the overall performance via controlling the CAV only. A similar idea has been recently proposed by Stern \etal~\cite{stern2018dissipation}, known as \emph{Lagrangian Control} of traffic flow, where CAVs are utilized as \emph{mobile actuators} to control the entire mixed traffic system. The effectiveness of this notion has been validated in various scenarios, including closed ring road~\cite{zheng2020smoothing}, open straight road~\cite{wang2020leading}, traffic bottleneck~\cite{vinitsky2018lagrangian}, and nonsignalized intersection~\cite{wu2017emergent}. To the best of our knowledge, the feasibility of this notion has not been explicitly discussed in the scenario of signalized intersections.
\end{remark}
\section{Methodology}
\label{Section:Methodology}
%{\color{red} Head paragraph. Stability and controllability are two fundamental properties of linear time invariant (LTI) systems. Thus,
Firstly we analyze the open-loop stability and controllability of the proposed ``$ 1+n $'' mixed platoon systems. Based on the derived stability and controllability conditions, an optimal control framework is proposed to optimize the CAVs' driving strategy in a single mixed platoon. Finally, an event-triggered algorithm is established to solve the collision problem of different mixed platoons.
%\begin{definition}
%At $t \in \mathbb{R}^{+}$, sum of all the mixed platoons in CZ is counted as $n(t) \in \mathbb{n}$. Let $\mathcal{n}(t)=\{1, \ldots, n(t)\}$ be the sequential set of all the mixed platoon, where the sequence is determined by the CAVs' entering time of CZ.
%\end{definition}
%Thus, the following equation is easy to be proved base on first-come-first-served principle:
%\begin{equation}
%\label{equ:time}
%t_{i}^{\rm f} \geq t_{i-1}^{\rm f}, \forall i \in \mathcal{n}(t), i>1
%\end{equation}
%In the equation, $ t_{i}^{\rm f} $ indicates when the leading CAV of mixed platoon $ k $ is planned to arrive at the stopping line(Divided line of MZ and CZ).
\subsection{Open-Loop Stability Analysis}
\label{Section:FundamentalProperties:Stability}
Based on the model~\eqref{equ:Platoon_state} of the mixed platoon system, we first consider its open-loop stability as shown in Definition~\ref{def:Stability}, when the leading CAV has no external control input, \ie, $u(t)=0$.
%\begin{definition}[Asymptotic Stability~\cite{skogestad2007multivariable}]
% \label{def:Stability}
% An LTI system $\dot{x} = Ax$ is said to be asymptotic stable if all the eigenvalues of $A$ have negative real parts.
%\end{definition}
{ \begin{definition}[Lyapunov Stability~\cite{skogestad2007multivariable}]
\label{def:Stability}
For a general dynamical system $\dot{x} = f(x(t))$, the equilibrium $x_e$ is said to be Lyapunov stable, if $\forall \epsilon>0$, there exists a $\delta >0$ such that, if $\Vert x(0) - x_e \Vert < \delta$, then for every $t\geq 0$ we have $\Vert x(t) - x_e \Vert < \epsilon$.
\end{definition}
Typically, the stability of a nonlinear system $\dot{x} = f(x(t))$ around an equilibrium point is analyzed after system linearization, and the linearized system $\dot{x} = Ax(t) $ is (asymptotically) stable if and only if all the eigenvalues of $A$ have negative real parts.} Existing research have revealed the stability criterion of the linearized car-following model~\eqref{equ:ErrorSpeedDiff} of one single HDV, shown in Lemma~\ref{equ2:Steady}.
\begin{lemma}[\cite{wilson2011car}]
\label{equ2:Steady}
The linearized car-following model~\eqref{equ:ErrorSpeedDiff} is stable if and only if $\alpha_{1}>0,\alpha_{2}>0$.
\end{lemma}
Regarding the ``$ 1+n $'' mixed platoon system, where there exist $n$ following HDVs, we have the following result.
%{\color{red} For simplicity, we use $ n $ to represent the $n$ following HDVs in the deduction.}
\begin{theorem}
\label{Th:Stability}
The ``$1+n$'' mixed platoon system~\eqref{equ:MixedPlatoon} is open-loop stable if and only if $\alpha_{1}>0, \alpha_{2}>0$, which is irrelevant to the platoon size $n$.
\end{theorem}
%\begin{comment}
%Another lemma~\ref{equ2:BlockMatrix} is the computation of block matrix, which will be used in the following deduction.
%
%\begin{lemma}
%\label{equ2:BlockMatrix}
%If matrix $ P = \left[ \begin{array}{cc}
%A & B \\
%C & D
%\end{array}\right]$, where $ P \in \mathbb{R}^{(m+n) \times(m+n)} $, $ A \in \mathbb{R}^{m \times m} $, $ D \in \mathbb{R}^{n \times n} $, $ B \in \mathbb{R}^{m \times n} $, $ C \in \mathbb{R}^{n \times m} $, if $ D $ is invertible, $ |P| = \left| \begin{array}{cc}
%A & B \\
%C & D
%\end{array}\right| = \left|A-BD^{-1}C\right|\left|D\right| $.
%\end{lemma}
%\begin{IEEEproof}
%If $ D $ is invertible, there exist
%\begin{equation}
%\left[\begin{array}{cc}
%A & B \\
%C & D
%\end{array}\right] \left[\begin{array}{cc}
%I_m & 0 \\
%-D^{-1}C & I_n
%\end{array}\right] = \left[\begin{array}{cc}
%A - BD^{-1}C & B \\
%0 & D
%\end{array}\right],
%\end{equation}
%calculate the determinant of both side of the equation, we have $ |P| = \left| \begin{array}{cc}
%A & B \\
%C & D
%\end{array}\right| = \left|A-BD^{-1}C\right|\left|D\right| $.
%\end{IEEEproof}
%\end{comment}
{The proof of Theorem~\ref{Th:Stability} is shown in~\ref{sec:Appendix}. Note that there always exist two zero eigenvalues in $A$, as shown in~\eqref{equ2:n2:EigenValue}, and thus the mixed platoon system~\eqref{equ:MixedPlatoon} is Lyapunov stable, but not asymptotically stable. It can be easily seen that the zero eigenvalues are brought by the states of the leading CAV in the open-loop case. Indeed, when $\alpha_{1}>0, \alpha_{2}>0$, the subsystem consisting of the states of the following HDVs is strictly asymptotically stable, while the mixed platoon system is Lyapunov stable.}
{Recall that the specific value of the linear coefficients $\alpha_i \,(i=1,2,3)$ under the OVM model~\eqref{equ2:OVM_a} has been derived in~\eqref{equ2:OVM_vopt_alphai}. According to Theorem~\ref{Th:Stability}, it is straightforward to obtain the stability condition for the OVM model after linearization: $ {\kappa} > 0, V_2C_1 > 0 $. Note that string stability is also an important topic for CAV's longitudinal control, which describes the propagation of the perturbations in a string of vehicles. Existing research for string stability of mixed platoons mostly focuses on one special case where one CAV is following at the tail behind a string of HDVs; see, \ie, \cite{wu2018stabilizing,jin2014dynamics}. Interested readers are referred to~\cite{wang2020leadingSCS} for further analysis on the proposed ``$ 1+n $" mixed platoon system and more general cases.}
\subsection{Controllability Analysis}
\label{Section:FundamentalProperties:Controllability}
%{\color{red} Needs to mention the formal definition and significance of controllability. Also state the result as a formal theorem. Refer to the IoTJ paper~\cite{zheng2020smoothing} or the TITS paper~\cite{wang2020controllability}.}
One goal of the control of the mixed platoon is to pass the intersection with a pre-specified equilibrium velocity, and controllability is a fundamental property to depict the feasibility of this goal. Particularly, if controllability holds for the mixed platoon, then the mixed platoon system can be moved to any desired state under the control input of the CAV. The formal definition and one useful criterion of controllability are shown as follows.
%If a system is controllable, its closed-loop poles can be placed arbitrarily.
\begin{definition}[Controllability~\cite{skogestad2007multivariable}]
\label{def:Controllability}
The dynamical system $ \dot{x}(t)=Ax(t)+Bu(t) $, or the pair $ (A,B) $, is controllable if and only if, for any initial state $ x(0)=x_0 $, any time $ t_f>0 $ and any final state $ x_f $, there exists an input $ u(t) $ such that $ x(t_f)=x_f $.
\end{definition}
\begin{lemma}[Popov-Belevitch-Hautus criterion~\cite{skogestad2007multivariable}]
\label{equ2:Control:PBH}
In a continuous-time LTI system $({A},{B})$ of size $n$, the system is controllable if and only if for every eigenvalue $ \lambda $, $ {\rm rank}\left({\lambda}{I} - {A},{B}\right) = n$.
System $({A},{B})$ is uncontrollable if and only if there exists $ {\rho} \neq 0$, such that ${\rho}^{\top} {A}=\lambda {\rho}^{\top}$, ${\rho}^{\top} {B}=0$.
\end{lemma}
Our result regarding the controllability of the $1+n$ mixed platoon is as follows.
\begin{theorem}
\label{Th:Controllability}
The ``$1+n$'' mixed platoon system~\eqref{equ:MixedPlatoon} is controllable when the following condition holds, which is irrelevant to the platoon size $n$.
\begin{equation}
\label{equ:ControllabilityCondition}
\alpha_{1}-\alpha_{2} \alpha_{3}+\alpha_{3}^{2} \neq 0.
\end{equation}
\end{theorem}
\begin{proof}
Assume that the mixed platoon system~\eqref{equ:MixedPlatoon} is uncontrollable. According to Lemma~\ref{equ2:Control:PBH}, there exists a scalar $\lambda$ and a non-zero vector $ {\rho} = [\rho_{01},\rho_{02},\rho_{11},\rho_{12},\dots,\rho_{n1},\rho_{n2}]^ {\rm T} $, where $ \rho_{ij}\in \mathbb{R} $, which satisfy
\begin{equation}
\rho^ {\rm T}({A}-\lambda {I})=0, \; \rho^ {\rm T}{B} = 0.
\end{equation}
From $ \rho^ {\rm T}{B} = 0 $, we have $ \rho_{02} = 0 $. From $\rho^ {\rm T}({A}-\lambda {I})=0$, it is obtained that
\begin{equation}
\label{equ2:1:EigenVector}
\begin{cases}
{-\lambda \rho_{01}=0}, \\
{\rho_{01}-\lambda \rho_{02}+\rho_{11}+\alpha_{3} \rho_{12}=0},
\end{cases}
\end{equation}
and
\begin{equation}
\label{equ2:n:EigenVector}
-\rho_{n1}-\left(\alpha_{2}+\lambda\right) \rho_{n2}=0,
\end{equation}
and for $i=1,\ldots,n$,
\begin{equation}
\label{equ2:1_n:EigenVector}
\begin{cases}
{-\lambda \rho_{i1}+\alpha_{1} \rho_{i2}=0}, \\
{\rho_{i1}+\left(-\alpha_{2}-\lambda\right) \rho_{i2}+\rho_{(i+1)1}+\alpha_{3} \rho_{(i+1)2}=0}.
\end{cases}
\end{equation}
According to~\eqref{equ2:1_n:EigenVector}, we have
\begin{equation}
\label{equ2:After:i}
\left(\lambda^{2}+\alpha_{2}\lambda+\alpha_{1}\right) \rho_{i1}=\left(\alpha_{3}\lambda+\alpha_{1}\right) \rho_{(i+1)1}, \; i=1,\ldots,n,
\end{equation}
and when $ i = 1 $, it holds that $ \lambda\rho_{11} = \alpha_{1}\rho_{12} $. Substituting it into~\eqref{equ2:1:EigenVector} yields
\begin{equation}
\label{equ2:After:1}
\left(\alpha_{3} \lambda+\alpha_{1}\right) \rho_{11}=0.
\end{equation}
From~\eqref{equ2:n:EigenVector}, we have
\begin{equation}
\label{equ2:After:n}
\left(\lambda^{2}+\alpha_{2} \lambda+\alpha_{1}\right) \rho_{n 1}=0.
\end{equation}
It can be easily examined that when $\alpha_{1}-\alpha_{2} \alpha_{3}+\alpha_{3}^{2} \neq 0$, the two equations
$\lambda^{2}+\alpha_{2} \lambda+\alpha_{1} = 0 $ and $ \alpha_{3}{\lambda}+\alpha_{1} = 0 $ cannot be satisfied simultaneously. Therefore, we consider the following three cases: (1)${\lambda}^{2}+\alpha_{2}\lambda+\alpha_{1} = 0$, $\alpha_{3}{\lambda}+\alpha_{1} \neq 0$; (2)${\lambda}^{2}+\alpha_{2}\lambda+\alpha_{1} \neq 0$, $\alpha_{3}{\lambda}+\alpha_{1} = 0$; (3)${\lambda}^{2}+\alpha_{2}\lambda+\alpha_{1} \neq 0$, $\alpha_{3}{\lambda}+\alpha_{1} \neq 0$. In each case, it can be obtained that $ \rho_{i1}=\rho_{i2}=0 $, $i=0,1,\ldots,n$ by combing $ \rho_{02} = 0 $ and equations~\eqref{equ2:After:i},~\eqref{equ2:After:1},~\eqref{equ2:After:n}. This contradicts the requirement that $ \rho \neq 0 $, which indicates that the assumption does not hold. Therefore, the system $({A},{B})$ is controllable when condition~\eqref{equ:ControllabilityCondition} holds.
\end{proof}
\begin{remark}
Theorem~\ref{Th:Controllability} indicates that the ``$1+n$'' mixed platoon is controllable with regard to the control input of the leading CAV when condition~\eqref{equ:ControllabilityCondition} holds. This result indicates that through controlling the leading CAV directly, one has complete control of the motion of the following $n$ HDVs without changing their natural driving behaviors. This property allows for the feasibility of designing the control input of the single CAV with the aim of improving the performance of the entire ``$1+n$'' mixed platoon. Note that condition~\eqref{equ:ControllabilityCondition} is consistent with previous controllability analysis in~\cite{zheng2020smoothing,cui2017stabilizing}, which focused on mixed platoons in a closed ring-road traffic system. Interested readers are referred to~\cite{wang2020controllability} to make further investigations on the controllability property of the ``$1+n$'' mixed platoon when the following $n$ HDVs have heterogeneous dynamics in~\eqref{equ:CFmodel}.
\end{remark}
\subsection{Optimal Control Framework}
\label{Section:OptimalControlMode}
After the dynamical analysis of the fundamental properties of the proposed ``$ 1+n$'' mixed platoon, in the following section, we proceed to establish an optimal control framework for the mixed platoon at the signalized intersection.
%Constraints. Performance includes traffic efficiency and fuel consumption.
%
%In this section, we present {\color{red}the optimal control framework of the mixed platoon system when passing an intersection}.
%Car-following model used in the modeling is firstly introduced. And cost function is the most crucial design, so the terminal velocity is optimized to a proper value, rather than the highest velocity. The constrains and the overall optimal control are interpreted at the end of this section.
%\subsection{Car-following Model}
%For simplicity, we rewrite the system function. Second order dynamic model is used to describe the vehicles: {\color{red} This could be partly moved to previous section and Section ``Terminal Speed''.}
%
%\begin{equation}
%\label{equ2:n+1_vehicle_state}
%\begin{split}
%\boldsymbol{X}
%&=\left[
%\begin{matrix}
%\boldsymbol{x_0} & \boldsymbol{x_1} \dots & \boldsymbol{x_n}
%\end{matrix}
%\right]^ {\rm T}\\
%&=\left[
%\begin{matrix}
%x_0 & v_0 & x_1 & v_1 \dots & x_\mathrm{n} & v_\mathrm{n}
%\end{matrix}
%\right]^ {\rm T}
%\end{split}
%\end{equation}
%\begin{equation}
%\label{equ2:n+1_vehicle_dynamic}
%\begin{split}
%\dot{\boldsymbol{X}}
%&=\left[
%\begin{matrix}
%\dot{\boldsymbol{x}}_{\boldsymbol{0}} & \dot{\boldsymbol{x}}_{\boldsymbol{1}} \dots & \dot{\boldsymbol{x}}_{\boldsymbol{n}}
%\end{matrix}
%\right]^ {\rm T}\\
%&=\left[
%\begin{matrix}
%v_0 & a_0 & v_1 & a_1 \dots & v_\mathrm{n} & a_\mathrm{n}
%\end{matrix}
%\right]^ {\rm T}
%\end{split}
%\end{equation}
%
%\begin{equation}
%\label{equ2:n+1_control}
%\boldsymbol{U_i}=
%\left[
%\begin{matrix}
%0 & u_0 & 0 & u_1 & \dots & 0 & u_\mathrm{n}
%\end{matrix}
%\right]^ {\rm T}
%\end{equation}
\subsubsection{Cost Function}
In our optimal control framework of the ``$ 1+n$'' mixed platoon, the main control objective is to let the CAV reach the stopping line of the intersection when the traffic signal turns green, and meanwhile the following HDVs are stabilized at a desired equilibrium velocity $v^{\ast}$, as discussed in~\eqref{equ:CFmodel:Stable}. %Note that the headway distance of $ n $ HDVs is also expected to be the corresponding distance $ d^{\ast} $.
Moreover, we also aim at minimizing the fuel consumption of the entire mixed platoon during its process of approaching the intersection. Accordingly, we define the following cost function in the Bolza form
\begin{equation}
\label{equ2:cost_fuction}
\newcommand{\ud}{\mathrm{d}}
J = \varphi({X}(t_\mathrm{f})) + \int_{t_0}^{t_\mathrm{f}}L({X}(t),{u}(t))\ud t,
\end{equation}
where $ t_0 $ is the time when CAV enters CZ, \ie, reaches the boundary of CZ and OZ as shown in Fig.~\ref{fig:Algorithm}. And $t_\mathrm{f}$ denote the time when CAV enters CZ, \ie, reaches the stopping line, which will be discussed later in Section~\ref{Section:Algorithm:MixedPlatoon}.
%where $t_0$ and $t_f$ denote the time when the CAV enters CZ, and the time when the CAV enters MZ, respectively.
As the terminal cost function in~\eqref{equ2:cost_fuction}, $\varphi({X}(t_f))$ measures the deviation of the system final state from the desired state, which is defined as
\begin{equation}
\label{equ2:ternimal_cost_fuction}
\varphi\big({X}(t_\mathrm{f})\big) = \omega_1\big(x_0(t_\mathrm{f}\big) - x_{\mathrm{tar}})^2 + \omega_2\sum_{i=0}^{n}\big(v_i(t_\mathrm{f}) - v^{\ast}\big)^2,
\end{equation}
where $ \omega_1 $ and $ \omega_2 $ denote the weight coefficients for penalty of the position deviation of the leading CAV and the velocity deviation of all the vehicles in the mixed platoon, respectively. $ x_0(t_\mathrm{f}) $ denotes the position of the leading CAV at $ t=t_\mathrm{f} $. $ x_{\mathrm{tar}} $ is the target final position of the leading CAV, which refers to the position of the stopping line at the intersection. The specific choice of the desired equilibrium velocity $ v^{\ast} $ and the target position $ x_{\mathrm{tar}} $ of the CAV will be discussed in Sections~\ref{Sec:Terminal Speed} and~\ref{Sec:Constraints}, respectively.
In~\eqref{equ2:cost_fuction}, $L\big({X}(t),{u}(t)\big)$ denotes the transient fuel consumption of the mixed platoon at time $t$, which is defined as
\begin{equation}
\label{equ2:Fuel_consumption_all}
L\big({X}(t), {u(t)}\big)=G_{0}(t)+\sum_{i = 1}^{n} G_{i}(t),
\end{equation}
where $G_{0}(t)$ and $G_{i}(t)$, $i=1,\ldots,n$ represent the transient fuel consumption of the leading CAV and the following HDVs, respectively. Similar to recent work~\cite{zhao2018platoon, jiang2017eco}, we utilize the Akcelik's fuel consumption model for the specific model to calculate transient fuel consumption~\cite{akcelik1989efficiency}
\begin{equation}
\label{equ2:Akcelik}
G_{i}(t)=\alpha+\beta_{1}P_{T}(t)+\left(\beta_{2} m a_{i}(t)^{2} v_{i}(t)\right)_{a_{i}(t)>0},
\end{equation}
where $ m $ is the vehicle mass, and the term $ \left(\beta_{2} m a_{i}(t)^{2} v_{i}(t)\right)_{a_{i}(t)>0} $ represents the extra inertial (engine/internal) drag power in vehicle acceleration. $\alpha$ is the idle fuel consumption rate and $P_{T}$ denotes the total power to drive the vehicle, which contains the engine dragging power, moment of inertia, air friction and other energy loss; it can be computed by
\begin{equation}
\label{equ2:Akcelik:Power}
P_{T}(t)=\max \left\{0, d_{1} v_{i}(t)+d_{2} v_{i}(t)^{2}+d_{3} v_{i}(t)^{3}+ma_{i}(t)v_{i}(t)\right\}.
\end{equation}
As suggested in~\cite{akcelik1989efficiency}, we consider a typical setup for parameter values in the Akcelik's fuel consumption model~\eqref{equ2:Akcelik} and~\eqref{equ2:Akcelik:Power}, as shown in Table~\ref{tab:Akcelik}.
\begin{remark}
% {\color{red}Zhao introduced a practical method to solve the problem by optimal control~\cite{zhao2018platoon}. We use the same optimal control idea in our modeling. Differently, we explicit design only one leading CAV in the mixed platoon as shown in~\eqref{equ:MixedPlatoon}, and improve the terminal velocity setup as shown in~\eqref{equ2:ternimal_cost_fuction}.}
Note that the minimization of fuel consumption is one typical control objective for control of individual vehicles at the intersection, which is known as the \emph{eco-approaching behavior}~\cite{yang2017eco,li2015eco, jiang2017eco}. However, existing research mostly focused on the behaviors of the CAVs themselves; such consideration might limit the potential of CAVs in improving traffic performance, especially in mixed traffic flow where HDVs also exist. One of the major distinctions in our optimal control framework from previous results~\cite{yang2017eco,li2015eco, jiang2017eco} lies in the explicit consideration of the fuel consumption of both CAVs and HDVs. This framework allows one to improve the fuel economy of the entire mixed traffic intersection via direct control of only CAVs.
\end{remark}
\begin{table}[t]
\centering
\begin{tabular}[c]{cccccccc}
\toprule%[1.5pt]
\multirow{2}{0.09\textwidth}{\centering\textbf{Parameter}} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$\bm{\alpha}$\\($\bm{\mathrm{ml/s}}$)}} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$\bm{\beta_{1}}$}\\} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$\bm{\beta_{2}}$}\\} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$ \bm{m} $\\($\bm{\mathrm{kg}}$)}} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$\bm{d_{1}}$}\\} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$\bm{d_{2}}$}\\} &
\multirow{2}{0.09\textwidth}{\centering\textbf{$\bm{d_{3}} $}\\}\\
& & & & & & & \\
\midrule%[1pt]
Value & 0.666 & 0.072 & 0.0344 & 1680 & 0.269 & 0.0171 & 0.000672 \\
\bottomrule%[1.5pt]
\end{tabular}
\caption{Parameter setup in the fuel consumption model~\eqref{equ2:Akcelik} and~\eqref{equ2:Akcelik:Power}}
\label{tab:Akcelik}
\end{table}
%\begin{table}[t]
% \centering
% \caption{Fuel Consumption Parameters}
% \label{tab:Akcelik}
% \begin{tabular}[c]{cccccc}
% \toprule%[1.5pt]
% {Parameter} & {Value} & {Unit} & {Parameter} & {Value} & {Unit}\\\midrule%[1pt]
% $\alpha$ & 0.666 & ml/s & $d_{1}$ & 0.269 &\\
% $\beta_{1}$ & 0.072 & & $d_{2}$ & 0.0171 & \\
% $\beta_{2}$ & 0.0344 & & $d_{3} $ & 0.000672 & \\
% $ m $ & 1680 & kg & & &\\
% \bottomrule%[1.5pt]
% \end{tabular}
%\end{table}
\begin{table}[t]
\centering
\begin{tabular}[c]{ccccccc}
\toprule%[1.5pt]
\multirow{2}{0.11\textwidth}{\centering\textbf{Parameter}} &
\multirow{2}{0.11\textwidth}{\centering\textbf{$\bm{\kappa}$\\($\bm{\mathrm{s^{-1}}}$)}} &
\multirow{2}{0.11\textwidth}{\centering\textbf{$\bm{V_1}$\\($\bm{\mathrm{m/s}}$)}} &
\multirow{2}{0.11\textwidth}{\centering\textbf{$\bm{V_2}$\\($\bm{\mathrm{m/s}}$)}} &
\multirow{2}{0.11\textwidth}{\centering\textbf{$ \bm{C_1} $\\($\bm{\mathrm{m^{-1}}}$)}} &
\multirow{2}{0.11\textwidth}{\centering\textbf{$\bm{C_2}$}\\} &
\multirow{2}{0.11\textwidth}{\centering\textbf{$\bm{L_\mathrm{veh}}$\\($\bm{\mathrm{m}}$)}}\\
& & & & & & \\
\midrule%[1pt]
Value & 0.85 & 6.75 & 7.91 & 0.13 & 1.57 & 5 \\
\bottomrule%[1.5pt]
\end{tabular}
\caption{Parameter setup in the OVM model~\eqref{equ2:OVM_a} and~\eqref{equ2:OVM_vopt}}
\label{tab:OVM-parameter}
\end{table}
%\begin{table}[t]
% \centering
% \caption{OVM Parameters}
% \label{tab:OVM-parameter2}
% \begin{tabular}[c]{cccccc}
% \toprule%[1.5pt]
% {Parameters} & {Value} & {Unit} & {Parameters} & {Value} & {Unit}\\\midrule%[1pt]
% $\kappa$ & 0.85 & s$^{-1}$ & $C_1$ & 0.13 & m$^{-1}$ \\
% $V_1$ & 6.75 & m/s & $C_2$ & 1.57 & \\
% $V_2$ & 7.91 & m/s & $L_\mathrm{veh}$ & 5 & m\\
% \bottomrule%[1.5pt]
% \end{tabular}
%\end{table}
\subsubsection{Terminal Velocity}
\label{Sec:Terminal Speed}
\begin{figure*}[t]
\centering
\subfigure[\label{fig:DistanceNumber:a}] {\includegraphics[width=0.4\linewidth]{Speed2Distance-eps-converted-to.pdf}}
\hfil
\subfigure[\label{fig:DistanceNumber:b}] {\includegraphics[width=0.4\linewidth]{Speed2Numbers-eps-converted-to.pdf}}
\caption{{Relationship between equilibrium velocity of the mixed platoon (\ie, terminal platoon velocity) and passing number $n$ of HDVs. OVM is taken as the car-following model, whose parameters are shown in Table~\ref{tab:OVM-parameter}. With the increase of equilibrium velocity, the equilibrium headway distance of HDVs increases, while the passing number of HDVs first rises and then drops. Hence, for constant traffic SPAT, there exists a maximum value of equilibrium velocity that maximizes the passing number of HDVs, highlighted as the red point.} \label{fig:DistanceNumber}}
\end{figure*}
We proceed to discuss how to design the desired equilibrium velocity $v^*$, which also represents the terminal velocity in the terminal cost function~\eqref{equ2:ternimal_cost_fuction}. Existing research mostly focused on the control of the CAV alone, and thus they typically set the terminal velocity of the CAV as the highest limited velocity in order to improve traffic efficiency at the intersection; see, \eg,~\cite{asadi2011predictive,jiang2017eco}. Considering that there might exist other HDVs at the intersection, we reveal that this setup in previous works might not be the optimal choice for the entire mixed intersection.
When designing the terminal velocity, we aim at maximizing the number of vehicles that can pass the intersection in an equilibrium state during a constant green phase time $ T_{\mathrm{Green}} $.
%We consider the traffic light as constant phase time traffic light, whose information is available to the CAVs by V2I technology. We claim that
Take one ``$ 1+n$'' mixed platoon for example, \ie, there is one leading CAV and $ n $ following HDVs in the platoon. From the equilibrium equation~\eqref{equ:CFmodel:Stable} in the HDVs' car-following model, it can be inferred that the equilibrium headway distance $ d^* $ relies on the equilibrium velocity $ v^* $. For constant green light phase time, our optimization goal is to maximize the number $ n $ of the following HDVs in $ T_{\mathrm{Green}} $. Accordingly, we have the following result.
\begin{definition}[Optimal velocity $ v^{*} $]
\label{Th:TerminalVelocity}
Consider the ``$ 1+n $'' mixed platoon system consisting of one leading CAV and $n$ following HDVs given by~\eqref{equ:MixedPlatoon}. The optimal equilibrium velocity $ v^{*} $, \ie, the optimal target velocity, is defined as the equilibrium velocity of the mixed platoon which maximizes the passing number $ n $ of the following HDVs during a constant green phase time $ T_{\mathrm{Green}} $. It can be obtained by solving the following optimization problem:
\begin{equation}
\label{equ2:T_Green:N}
\begin{aligned}
&\arg\max_{v^*} &n = \frac{v^* T_{\mathrm{Green}}}{d^*}, \\
&\mathrm{subject\; to: } &F(d^*,0,v^*)=0.
\end{aligned}
\end{equation}
\end{definition}
{Recall that we utilize the OVM model~\eqref{equ2:OVM_a} to derive the explicit expression of HDVs' car-following model $F(\cdot)$, with the specific value of the linearized coefficients in~\eqref{equ:ErrorSpeedDiff} calculated by~\eqref{equ2:OVM_vopt_alphai}. Motivated by~\cite{helbing1998generalized}, we consider a typical parameter setup of the OVM model as shown in Table~\ref{tab:OVM-parameter}. Note that both open-loop stability and controllability hold in this parameter setup.} %, as discussed in Section~\ref{Section:FundamentalProperties:Stability} and \ref{Section:FundamentalProperties:Controllability}.
{The leading CAV is expected to reach the stopping line when the traffic light turns green.} Accordingly, if the mixed platoon is in equilibrium state, the leading CAV and following HDVs run at the same velocity, \ie, $F(\cdot)= \dot{v}_{i} = 0$ holds for the following HDVs. Thus, we obtain the relationship between velocity and headway distance in equilibrium state based on the OVM model as follows. Substituting $F(\cdot)= \dot{v}_{i} = 0$ into equation~\eqref{equ2:OVM_vopt} yields
\begin{equation}
\label{equ2:Steady_distance}
d_i = \frac{1}{C_1}\left(\arctan\left(\frac{v_i - V_1}{V_2}\right) + C_2\right) + L_\mathrm{veh}.
\end{equation}
{In Fig.~\ref{fig:DistanceNumber}, it can be observed that the equilibrium headway distance of HDVs is typically a monotonically increasing function with respect to equilibrium velocity in the equilibrium state. There exists a maximum passing number of HDVs corresponding to the optimal equilibrium velocity $ v^{*} $ and equilibrium headway distance $ d^{*} $. The optimal terminal velocity $ v^{*} $ can be obtained by solving \eqref{equ2:T_Green:N}.}
%{\color{red}By calculating the partial derivative of $n(v)$, the optimal terminal velocity $ v^{*} $ in \eqref{equ2:T_Green:N} can be obtained. In Fig.~\ref{fig:DistanceNumber}, it can be clearly seen that because the car-following distance is not a linear function to vehicle velocity. However, mixed platoon travel distance is proportional to the platoon velocity in constant green phase $ T_{\mathrm{Green}} $, thus there exists a maximum amount of passing vehicles corresponding to the optimal equilibrium velocity $ v^{*} $ and equilibrium headway distance $ d^{*} $.}
{
\begin{remark}
We make further analysis on the optimal velocity $v^*$. If $ F(d^*,0,v^*)=0 $ leads to an explicit expression of $ d^* = d(v^*) $, solving~\eqref{equ2:T_Green:N} yields
\begin{equation}
\frac{\partial}{\partial{v^*}}\left(\frac{v^*T_{\mathrm{Green}}}{d(v^*)}\right) = 0,
\end{equation}
which leads to
\begin{equation}
\label{equ:vstar_Tgreen}
d(v^*) - d'(v^*)v^* = 0.
\end{equation}
From \eqref{equ:vstar_Tgreen} it can be observed that the optimal velocity $ v^* $ is only related to the car-following model and its equilibrium equation $ F(d^*,0,v^*)=0 $. It has no relationship from the signal phasing time $ T_{\mathrm{Green}} $. By contrast, the maximum passing number $ n $ depends on the value of both $ T_{\mathrm{Green}} $ and $ v^* $.
\end{remark}
}
\begin{remark}
Our result is consistent with the typical observations from the perspective of macroscopic traffic theory. Denote $\rho(x,t),v(x,t),q(x,t)$ as the traffic density, traffic flow velocity and traffic flow volume at position $x$ and time $t$, respectively. {The fundamental Lighthill-Whitham-Richard model~\cite{lighthill1955kinematic} is commonly employed to depict the relationship among them, which is shown in~\eqref{equ2:LWR} and~\eqref{equ2:LWR_flow}
\begin{equation}
\label{equ2:LWR}
\frac{\partial\rho(x,t)}{{\partial}t} + \frac{{\partial}q(x,t)}{{\partial}x} = 0,
\end{equation}
\begin{equation}
\label{equ2:LWR_flow}
q(x,t) = Q(\rho(x,t)),
\end{equation}}
where $ Q $ is usually a convex and non-monotonic function---typical results reveal that the traffic flow volume $q$ usually grows up first and then drops down as the increase of the traffic density $\rho$. Similarly, our result suggest that the optimal target velocity for the mixed platoon is not "the higher, the better".
\end{remark}
\subsubsection{Constraints}
\label{Sec:Constraints}
For practical implementation of the obtained controller of the CAV, there also exist several constraints that need to be taken into consideration, including process constraints and terminal constraints. Regarding process constraints, first is the safety constraint, which means that each vehicle in the mixed platoon should keep a safe distance $d_{\mathrm{safe}}$ from the preceding vehicle.
\begin{equation}
\label{equ2:Constraint_Distance}
x_{i}(t) - x_{i-1}(t) - L_{\mathrm{veh}} \ge d_{\mathrm{safe}},\, \mathrm{for}\;t_0 \leq t \leq t_{\mathrm{f}},\,i=1,2,\ldots,n.
\end{equation}
Second is the practical constraint of the value of the velocity and acceleration of each vehicle in the mixed platoon. Denote $v_{\mathrm{max}}$ as the maximum velocity, $a_{\mathrm{min}}$ and $a_{\mathrm{max}}$ as the minimum and maximum acceleration, respectively. Then, it should be satisfied that
\begin{align}
0 \le v_{i}(t) \le v_{\max},\, &\mathrm{for}\;t_0 \leq t \leq t_{\mathrm{f}},\,i=0,1,2,\ldots,n ; \label{equ2:Constraint_Speed}\\
a_{\min} \le {a}_{i}(t) {\le}a_{\max},\, &\mathrm{for}\;t_0 \leq t \leq t_{\mathrm{f}},\,i=0,1,2,\ldots,n. \label{equ2:Constraint_Acceleration}
\end{align}
{For the terminal constraint, we mainly focus on the terminal position of the CAV. Recall that the deviation of the terminal position $ x_{0}(t_{\mathrm{f}}) $ of the CAV from the target position $x_{\mathrm{tar}}$ has been penalized in the terminal cost function~\eqref{equ2:ternimal_cost_fuction}. Here, we further add an inequality constraint to require the CAV to neither pass the stopping line nor leave a large spacing away from it, given as follows
\begin{equation}
\label{equ2:Constraint_Ternimal_Position_Relax}
0 \le x_{0}(t_{\mathrm{f}}) \le x_{0}^{\max}(t_{\mathrm{f}}),
\end{equation}
where $x_{0}^{\max}$ denotes the maximum tolerance spacing of the CAV from the stopping line. Note that most single CAV control algorithms (\eg,~\cite{asadi2011predictive}) would not incorporate the penalty for the terminal position $ x_{0}(t_{\mathrm{f}}) $ in the terminal cost. Instead, they typically consider a hard constraint to require $ x_{0}(t_{\mathrm{f}}) = 0 $. These algorithms are mostly suitable for fully-autonomous scenarios. In our work for mixed traffic intersections, by contrast, the control of the ``$1+n$'' mixed platoon aims at passing the intersection with an optimal velocity $ v^{*} $ and meanwhile reducing fuel consumption. If we predetermine a fixed setup for both terminal time $ t_\mathrm{f} $ and terminal position $ x_{0}(t_{\mathrm{f}}) $ similar to those single CAV control algorithms, the feasible region for the control input of the leading CAV could be greatly limited, which might jeopardize the optimal performance for the entire mixed platoon system.}
%However, it is difficult to limit the CAV's terminal position exactly at the stopping line because of the system modeling error. In fact, as long as CAV has not reached the stopping line at the terminal time step, running into the red light is avoided. So the original constraint can be relaxed.
%Considering these reasons, , which also means that the hard constraints~\eqref{equ2:Constraint_Ternimal_Position} is replaced into soft constraints~\eqref{equ2:Constraint_Ternimal_Position_Relax}. Although this transition waste a little green light phase, it is acceptable because the problem is much more likely to be feasible in the simulation.
\begin{remark}
Note that the process constraint~\eqref{equ2:Constraint_Distance} only focuses on the longitudinal position inside a ``$1+ n $'' mixed platoon. The collisions between two adjacent ``$1+ n $'' mixed platoons are not considered in the constraints in the optimal control formulation. To address the safety constraint between different mixed platoons, we introduce an event-triggered algorithm, which is presented in Section~\ref{Section:Algorithm}.
\end{remark}
\subsubsection{Optimal Control Formulation}
Lumping the aforementioned design of the cost function and the constraints, the overall optimal control problem can be formulated as follows
\begin{equation}
\newcommand{\ud}{\mathrm{d}}
\label{equ:OptimalControl}
\begin{aligned}
&\mathop{\arg\max}_{u(t)}\; J = \varphi({X}(t_\mathrm{f})) + \int_{t_0}^{t_\mathrm{f}}L({X}(t),{u}(t))\ud t, \\
&\mathrm{subject\; to: }\; \eqref{equ2:OVM_a}, \eqref{equ2:OVM_vopt}, \eqref{equ2:Constraint_Distance}, \eqref{equ2:Constraint_Speed}, \eqref{equ2:Constraint_Acceleration},\eqref{equ2:Constraint_Ternimal_Position_Relax}, \\
&\mathrm{given: }\; {X}(t_\mathrm{0}).
\end{aligned}
\end{equation}
\begin{table}[t]
\centering
\begin{tabular}[c]{cc}
\toprule%[1.5pt]
{\textbf{Parameter}} & {\textbf{Value}} \\
\midrule%[1pt]
setup.nlp.solver & SNOPT \\
setup.nlp.snoptoptions.tolerance & $ 2 \times $10$^{-3}$ \\
setup.scales.method & automatic-bounds \\
setup.derivatives.derivativelevel & second \\
setup.mesh.tolerance & $ 10^{-2}$ \\
setup.mesh.iteration & $ 8 $ \\
setup.mesh.method & hp1 \\
setup.method & RPMintegration \\
\bottomrule%[1.5pt]
\end{tabular}
\caption{GPOPS Parameters}
\label{tab3:OptimalControlSolve}
\end{table}
Before solving Problem~\eqref{equ:OptimalControl}, the optimal terminal velocity $ v^{*} $ needs to be calculated first by solving Problem~\eqref{equ2:T_Green:N}. In addition, the terminal time $ t_\mathrm{f} $ needs to be pre-determined, which is discussed in the proposed algorithm in Section~\ref{Section:Algorithm}. To solve Problem~\eqref{equ:OptimalControl} numerically, which is a high-order nonlinear optimal control problem, we transform it into nonlinear programming (NLP) problem by employing the pseudo-spectral method~\cite{elnagar1995pseudospectral}. Several practical packages can be directly utilized to address this problem; see, \eg, the GPOPS (General-Purpose Optimal Control Software) toolbox~\cite{patterson2014gpops} with parameter setups shown in Table~\ref{tab3:OptimalControlSolve}. {Note that the optimal control problem can be formulated by utilizing any car-following model that satisfies~\eqref{equ:CFmodel}. In this paper we focus on OVM~\eqref{equ2:OVM_a}~\eqref{equ2:OVM_vopt} considering its balance between model fidelity and computational tractability. Indeed, the adopted pseudo-spectral method has its limitation when more complex nonlinear car-following models (\eg, IDM) are used. It is a significant future direction to seek for more efficient numerical methods to solve this high-order nonlinear optimal control problem.}
%The high-order nonlinear optimal control problem can not be solved in mathematical ways like variational method or maximum principle. Normally it is transformed into nonlinear programming (NLP) problem. The pseudo-spectral method is proposed by Elnagar, which is used in a practical package written by Patterson, namely . Some of the critical parameters is shown in Table~\ref{tab3:OptimalControlSolve}.
\subsection{Algorithm Design}
\label{Section:Algorithm}
In this section, we firstly introduce the benchmark algorithm for the control of CAVs at signalized intersections. Note that only the constraints in the mixed platoon are considered in Section~\ref{Sec:Constraints}. To avoid the potential collision between mixed platoons, we also present our event-triggered algorithm design for practical implementation of the ``$1+n$'' mixed platoon based method.
\begin{figure*}[!t]
\centering
\includegraphics[width=\linewidth]{Algorithm.pdf}
\caption{Mixed platoon algorithm. The left figure illustrates one arm of the intersection, which is separated into three zones corresponding to Section~\ref{section:ScenarioSetup}. {Note that the stopping line is set as the $ x=0 $ position and vehicle position $ x_i $ is calculated correspondingly.} The right figure illustrates the mixed platoon algorithm. CAV has four states. Before the CAV arrives at CZ, it is in \emph{uncontrolled} state. CAV becomes \emph{computed} state when it arrives at the boundary of OZ and CZ, where the optimization is performed. After entering CZ, CAV is in \emph{control} state and carries out the optimized velocity trajectory, unless CAV might collide with preceding vehicle thus it enters \emph{re-computed} state.}
\label{fig:Algorithm}
\end{figure*}
\subsubsection{Benchmark Algorithm}
\label{Sec:BenchmarkAlgorithm}
The basic benchmark algorithm is the predictive cruise control (PCC) algorithm proposed by Asadi \etal~\cite{asadi2011predictive}. {After obtaining traffic SPAT by V2I technology in advance, a practical method was proposed to choose the target green phase window, shown as follows
\begin{equation}
\label{equ3:PCC+}
[v_{\mathrm{low}},v_{\mathrm{high}}] = \left[\frac{D_{k}}{r_{j} - t}, \frac{D_{k}}{g_{j} - t}\right] \cap\left[v_{\min}, v_{\max}\right],
\end{equation}
where $ D_{k} $ is the distance from the CAV $ k $ to the stopping line; $ t $ is the current time; $ r_{j} $ is the start time of the next $ j $th red phase; $ g_{j} $ is that of the next $ j $th green phase; $ v_{\min} $ and $ v_{\max} $ are the CAV velocity limitations. Recall that the green light time in Definition~\ref{Th:TerminalVelocity} can be obtained as $ T_{\mathrm{Green}} = r_{j} - g_{j} $ where $ T_{\mathrm{Green}} > 0 $. The nonempty intersection $ [v_{\mathrm{low}},v_{\mathrm{high}}] $ shown in~\eqref{equ3:PCC+} indicates the feasible velocity window which allows the CAV to pass without idling. Then, the CAV obtains the target velocity $ v_{\mathrm{target}} = v_{\mathrm{high}} $, calculates the corresponding terminal time $ t_{\mathrm{f}} = \frac{D_{k}}{v_{\mathrm{target}}}$, and schedules an optimum velocity trajectory through model predictive control (MPC)~\cite{maciejowski2002predictive}.
}
%mainly focuses on how one CAV passes several sequential intersections}, which means it
However, PCC cannot consider preceding HDVs (especially the queuing ones) into optimization. Instead of PCC, we consider an improvement inspired by Yang's queuing length adjustment method~\cite{yang2017eco}. The distance from the CAV $ k $ to the stopping line $ D_{k} $ is optimized by
\begin{equation}
\label{equ3:PCC+:distance}
D_{k}^{*} =
\begin{aligned}
&\frac{v_k}{v_{k}+v_{\mathrm{AC}}}\left[D_{k}+v_{\mathrm{AC}}\left(r_{j}-t\right)\right], \\
\end{aligned}
\end{equation}
where $ v_{k} $ is the current velocity of CAV $ k $, and $ v_{\mathrm{AC}} $ is the velocity of the traffic upstream flow shock wave. In the rest of this paper, this modified PCC algorithm is named as PCC+.
%\begin{remark}
It can be inferred that even if the queuing length is considered in the optimization by~\eqref{equ3:PCC+:distance}, the inner core of the PCC+ algorithm is still designed for one CAV alone, which makes it a qualified benchmark algorithm to make comparisons with the mixed platoon based algorithm. Note that the similar target time window chosen method~\eqref{equ3:PCC+} is employed in our algorithm.
%\end{remark}
%\subsubsection{Mixed Platoon Algorithm}
%\label{Section:Algorithm:Mixed}
%If CAV is in OZ, no optimization or control task is assigned to CAV. When each CAV arrives at CZ, its optimization consist of two steps:
%\begin{enumerate}[1.]
% \item Utilize the traffic light information to choose the target time window, \emph{i.e.}, the terminal arrive time at MZ.
% \item Set the terminal arrive time as the terminal constraint and optimize the vehicle's velocity trajectory.
%\end{enumerate}
\subsubsection{Algorithm Design for Mixed Platoon}
\label{Section:Algorithm:MixedPlatoon}
In this paper, we consider the traffic light as constant SPAT. We design a hierarchical mixed platoon based algorithm as shown in Fig.~\ref{fig:Algorithm}. In Section~\ref{Section:OptimalControlMode}, the velocity trajectory planning process has been explained in detail. However, there still remains one problem. In our expectation, the decision-making and trajectory planning processes are made when the CAV arrives at the border between OZ and CZ. But one-time planning cannot forecast the future trajectories of the preceding vehicles, which means there are still collision risks when the CAV is undertaking the planned velocity trajectory in CZ. To solve this problem, there are basically two solutions. One is called shooting heuristic~\cite{ma2017parsimonious}, which estimates the maximum possible trajectory boundaries of the preceding vehicle. The drawback is that the models' deviation makes it difficult to guarantee prediction accuracy. The other method is receding horizon~\cite{feng2015real}, which allows the CAV to make planning in a constant time period. However, this method brings a huge computation burden for real-time implementation.
% utilizing dynamic programming. The boundaries are further used as the path constraints in CAV's velocity trajectory planning.
In fact, trajectory interference seldom happens when the traffic flow density is not saturated. So distributing extra computation brings very limited benefit. Thus, we design an event-triggered method to guarantee safety with minimum calculation. The CAV state is separated into four states, \emph{uncontrolled}, \emph{computed}, \emph{controlled} and \emph{re-computed}.
%{\color{red}The following needs to be revised carefully for typos.}
%\subsubsection{\color{red}Uncontrolled}
\begin{enumerate}[(1)]
\item \emph{uncontrolled}
\end{enumerate}
{
As shown in the white boxes in Fig.~\ref{fig:Algorithm}, all the CAVs are in \emph{uncontrolled} state when driving in OZ, whose length is denoted as $ L_{\mathrm{obs}} $. During this period, all the CAVs will directly follow the HDVs' car-following model~\eqref{equ2:OVM_a} to control themselves. Meanwhile, all the vehicles, including HDVs and CAVs, are also allowed to change lanes freely in OZ, as highlighted in the assumptions in Section~\ref{section:ScenarioSetup}.
}
\begin{enumerate}[(2)]
\item \emph{computed}
\end{enumerate}
%\subsubsection{\color{red} Typo!}
When the CAV arrives at CZ, its state becomes \emph{computed}, as shown in the orange boxes in Fig.~\ref{fig:Algorithm}. At this time point, the CAV firstly receives the traffic SPAT and chooses its target green phase window. Based on the time window, the optimal velocity trajectory is calculated by optimal control framework as shown in Section~\ref{Section:OptimalControlMode}.
%Considering the acceleration constraints\eqref{equ2:Constraint_Acceleration} and velocity constraints\eqref{equ2:Constraint_Speed}, the target green phase should satisfies\eqref{equ3:PCC+}.
As mentioned before, the target window is also chosen by the method in~\eqref{equ3:PCC+}. Differently, only one CAV is considered in~\cite{asadi2011predictive}, and thus it's straightforward to set the target window as the closest one. In our research, however, there is an optimal terminal platoon velocity as designed in Section~\ref{Sec:Terminal Speed}. If the target velocity is set as the maximum velocity $ v_{\max} $, the optimization on the velocity trajectory becomes meaningless. {Therefore, the limitation on the $ v_{\max} $ \eqref{equ3:PCC+} is slacked to
\begin{equation}
\label{equ3:Asadi:2}
[v_{\mathrm{low}},v_{\mathrm{high}}] = \left[\frac{D_{k}}{r_{j} - t}, \frac{D_{k}}{g_{j} - t}\right] \bigcap\left[v_{\min }, \frac{v_{\max } + v^{*}}{2}\right],
\end{equation}
in which $ v^{*} $ means the optimal velocity calculated from~\eqref{equ2:T_Green:N}. Similar to the benchmark algorithm presented in Section~\ref{Sec:BenchmarkAlgorithm}, the target velocity is selected through the nonempty intersection $ [v_{\mathrm{low}},v_{\mathrm{high}}] $ with $ v_{\mathrm{target}} = v_{\mathrm{high}} $. The corresponding terminal time is calculated as $ t_{\mathrm{f}} = \frac{D_{k}}{v_{\mathrm{target}}}$, which is utilized as terminal time in the cost function~\eqref{equ2:cost_fuction} in the optimal control formulation.
}
{
\begin{remark}
Recall that a maximum value of the passing number $ n $ of vehicles is obtained when solving~\eqref{equ2:T_Green:N}. In practical traffic flow, however, the number of the following HDVs behind the leading CAV might not be exactly equal to $ n $. In particular, when MPR is high, there might be fewer HDVs than $ n $ behind the CAV, and there might be other CAVs except for the leading CAV in the ``$1+n$'' mixed platoon. In this case, our algorithm lets the other CAVs in the ``$1+n$'' mixed platoon to utilize the numerical expression of the HDV's car-following model (precisely, the OVM model~\eqref{equ2:OVM_a}\eqref{equ2:OVM_vopt}) to determine their control input. In this way, the optimal control formulation~\eqref{equ:OptimalControl} is preserved, which allows the design of the control input of the leading CAV. It would be an interesting future direction to apply cooperative control strategies to multiple CAVs in a mixed platoon when passing an intersection, which might further improve the intersection performance.
\end{remark}
}
%\subsubsection{Controlled}
\begin{enumerate}[(3)]
\item \emph{controlled}
\end{enumerate}
When the CAV is in CZ, it is in \emph{controlled} state, as the green boxes showed in Fig.~\ref{fig:Algorithm}. In this state, CAV executes the velocity trajectory planned in \emph{computed} state. If there is a preceding vehicle, the CAV simply checks the safety distance constraint~\eqref{equ2:Constraint_Distance} with the preceding vehicle in each step. If not, the CAV's state is changed into \emph{re-computed}.
%\end{enumerate}
\begin{enumerate}[(4)]
\item \emph{re-computed}
\end{enumerate}
{
In \emph{controlled} state, if the leading CAV is too close to the preceding vehicle (the threshold distance is set as $6\, \mathrm{m}$ in the simulation), it will become \emph{re-computed} state, which aims to slow down the CAV to avoid collision with the preceding vehicle. Specifically, if the distance to the stopping line is far enough for the CAV to do another optimization ($ D_i > k_{\mathrm{c}}L_{\mathrm{ctrl}} $, where $ L_{\mathrm{ctrl}} $ is the length of CZ and $ k_{\mathrm{c}} \in [0,1] $), it will re-compute its new optimal trajectory by following the same process in the \emph{computed} state. On the other hand, if the CAV is too close to the intersection ($ D_i \leq k_{\mathrm{c}}L_{\mathrm{ctrl}} $), there is little optimization space for the CAV, and thus in this case, the CAV will directly follow the HDVs' car-following model~\eqref{equ2:OVM_a} to pass the intersection in the remaining time.
%Note that there are two conditions where CAV is controlled by the car-following model. One is in \emph{uncontrolled} state, where CAV is accumulating mixed platoon in OZ. The other one is in \emph{re-computed} state when CAV is too close to the intersection and there exists no space to optimized again. In these two conditions, the collision-free for the CAV is guaranteed by the car-following model just like HDVs.
}
%In \emph{re-computed} state, the CAV is too close to the preceding vehicle, as the blue boxes showed in Fig.~\ref{fig:Algorithm}. At first, we should decide whether the remaining distance is long enough for the second optimization for CAV. We judge if the distance from CAV to stopping line is larger than $ k_{\mathrm{c}}L_{\mathrm{ctrl}} $, where $ L_{\mathrm{ctrl}} $ is the length of CZ, and $ k_{\mathrm{c}} \in [0,1] $. If yes, the CAV can be \emph{re-computed} state when the preceding vehicle drives away. If not, the CAV is switched to car-following model control.
\begin{table}[!t]
\centering
\begin{tabular}[c]{ccc}
\toprule%[1.5pt]
{\textbf{Parameters}} & {\textbf{Symbol}} & {\textbf{Value}} \\
\midrule%[1pt]
Simulation step (s) & $ T_{\mathrm{s}} $ & 0.1 \\
Maximum acceleration (m/s$^{2} $) & $ a_{\max} $ & 3 \\
Minimum acceleration (m/s$^{2} $) & $ a_{\min} $ & -6 \\
Terminal position weight & $ \omega_1 $ & $ 10^5 $ \\
Terminal velocity weight & $ \omega_2 $ & $ 10^4 $ \\
Maximum velocity (m/s) & $ v_{\max} $ & 15 \\
Minimum velocity (m/s)& $ v_{\min} $ & 0 \\
Control zone length (m) & $ L_{\mathrm{ctrl}} $ & 300 \\
Observation zone length (m) & $ L_{\mathrm{obs}} $ & 500\\
\bottomrule%[1.5pt]
\end{tabular}
\caption{Parameter setup for the traffic simulation}
\label{tab3:OptimalControlMode}
\end{table}
\section{Simulation Results and Discussion}
\label{Section:SimulationResult}
In this section, we evaluate the effectiveness of the proposed optimal control framework in Section~\ref{Section:OptimalControlMode} and the corresponding algorithm in Section~\ref{Section:Algorithm} based on large-scale traffic simulations. The nonlinear OVM model~\eqref{equ2:OVM_a} is utilized to model the dynamics of HDVs. Firstly, the simulation environment and evaluation index are introduced. Then, a case study under $50\%$ MPR is presented to analyze the algorithm performance. Finally, two simulation experiments are conducted, considering multiple traffic volumes and MPRs.
%To meet the joint-simulation requirement of SUMO and GPOPS, Matlab R2017b is selected as the basic simulation software.
\subsection{Simulation Environment and Evaluation Index}
%(Simulation of Urban MObility)
%is an open-source microscopic traffic flow simulation platform~{\color{red}[]}. {\color{red}After developed by Germany Institute of Transportation Systems, it
The traffic simulation is conducted in SUMO, which is widely used in traffic researches~\cite{lopez2018microscopic}. The simulation is run on Intel Core i7-7700 processor @3.6GHz. Some simulation parameters are shown in Table~\ref{tab3:OptimalControlMode}.
%\subsection{Terminal Speed Validation}
%Firstly, we want to verify that the optimal terminal velocity is not the highest velocity as explained in section~\ref{Sec:Terminal Speed}. In this case, we set the length of CZ as 300m and the green phase time as 20s. The total amount of the vehicles is 30. Each mixed platoon consists of 1 CAV in the front and 9 HDVs following it, so there are 3 mixed platoons. To avoid platoon merging's influence on the result, each platoon is forced to assigned to one independent green phase window.
%
%The simulation result is shown in Fig.~\ref{fig:TerminalSpeed}. Red real line is CAV's trajectory and black dot line is HDV trajectory.It can be concluded that with higher terminal velocity set to the optimization, the mixed platoons try to pass the intersection in higher velocity. However, the higher velocity also means higher car-following distance, which means higher occupied rate of the green phase. It can be inferred that green phase is not fully utilized with higher terminal velocity. What's more, because the optimization on the terminal velocity is on all of the vehicles, higher terminal velocity also means that the velocity fluctuation in the platoon at terminal time is also higher, which is also disadvantageous to traffic efficiency.
Denote $ t_{i}^{\mathrm{in}} $ as the time step when vehicle $ i $ enters CZ, and $ t_{i}^{\mathrm{out}} $ as the time step when it exits CZ. In this way, it means that the vehicle spends $ t_{i}^{\mathrm{out}} - t_{i}^{\mathrm{in}} $ time to travel through CZ, while it spends $ {L_{\mathrm{ctrl}}}/{v_{\max}} $ time to travel through CZ in free drive condition. Accordingly, Average Travel Time Delay (ATTD) is chosen to measure the average traffic efficiency for the mixed platoon, as shown in~\eqref{equ3:TravelTimeDelay}.
%The result is shown in Fig. It can be seen that the optimal is 9m/s rather than higher velocity.
\begin{equation}
\label{equ3:TravelTimeDelay}
\mathrm{ATTD} = \frac{1}{n} \sum_{i = 0}^{n} \left(t_{i}^{\mathrm{out}} - t_{i}^{\mathrm{in}}-\frac{L_{\mathrm{ctrl}}}{v_{\max}}\right).
\end{equation}
Meanwhile, fuel consumption per 100km is chosen as the evaluation index to measure the fuel economy. {We consider $ 100 $ vehicles in each experiment. Note that at the initial time of the simulation, there are no vehicles in the traffic network. Since our traffic optimization focuses on steady-state traffic flow, we exclude the performance of the firstly-entered part of the vehicles and focus on that of the last $ 50 $ vehicles in the performance evaluation for the entire intersection.}
%In addition, if we don't force the platoon to be aligned with the green phase start time, it's quite straight forward that the longer car-following distance brings less vehicle capacity in one green phase. It also means that some vehicles at the very end of the platoon maybe extruded into next green phase, which may even lower the traffic efficiency.
\subsection{A Case Study under 50\% MPR}
\begin{figure*}[!t]
\centering
\subfigure[No control input assigned to CAVs\label{fig:NC_700}] {\includegraphics[width=0.95\linewidth] {NoControl_Gapof_48-eps-converted-to.pdf}}
\subfigure[PCC+ control input assigned to CAVs\label{fig:OC_700}] {\includegraphics[width=0.95\linewidth] {OneControl_Gapof_48-eps-converted-to.pdf}}
\subfigure[Mixed platoon control input assigned to CAVs\label{fig:MC_700}] {\includegraphics[width=0.95\linewidth] {Gapof_48-eps-converted-to.pdf}}
\caption{{Vehicle trajectories of no control applied, PCC+ control and mixed platoon control in $ 750 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $ (MPR = 50\%). The solid lines represent the CAVs' trajectories, while the dashed dot lines represent HDVs' trajectories. Black solid lines mean there are no optimized control input assigned to CAV. Blue lines are for the PCC+ controlled CAVs while magenta lines are for the proposed mixed platoon control.} \label{fig:ControlCompare}}
\end{figure*}
We first conduct the simulation in $ 50\% $ MPR of $ 100 $ vehicles, which means there are randomly distributed $ 50 $ CAVs and $ 50 $ HDVs driving into the intersection. Then further comparisons are made between the proposed Mixed Platoon (MP) algorithm and the bench-marking algorithm PCC+ shown in Section~\ref{Section:Algorithm} at different traffic volumes.
%The purple dash lines are degraded-CAVs' trajectories, which are assigned car-following model just like HDVs as interpreted in Section~\ref{Section:Algorithm}.
%In Fig.~\ref{fig:NC_1000}$ \sim $\ref{fig:MC_1000}, the traffic volume is 1000 vehicle/(hour$ \cdot $lane). The blue dot line indicates the total travel time of the 100 vehicles. Considering the same total vehicle amount, if no control is assigned to CAVs, it takes 610s to let the traffic flow pass the intersection. With PCC+ applied, the total travel time is slightly reduced to around 600s. But with MP applied to CAVs, it is further reduced to 560s.
%
%In Fig.~\ref{fig:NC_800}$ \sim $\ref{fig:MC_800}, the traffic volume is 800 vehicle/(hour$ \cdot $lane). The yellow real line indicates the accumulation of the vehicles queuing before the intersection. It can be observed another benefit which is the accumulation rate of the queuing vehicles is delayed and flatted by CAVs' control. MP outperform PCC+ with further delaying the accumulation starting time from 210s to 300s.
In Fig.~\ref{fig:ControlCompare} there are three groups of vehicle trajectories in $ 750 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $, which represent no control input (colored in black), PCC+ control (colored in blue) and proposed mixed platoon control (colored in magenta), respectively. Note that vehicle distributions in three scenarios are exactly the same. It can be observed that in Fig.~\ref{fig:NC_700} all the vehicles are queuing because of the red light, which causes vehicles' idling and fuel waste. In Fig.~\ref{fig:OC_700} the queuing is postponed but still gradually accumulating. On the contrary, with MP applied, the queuing accumulation is constrained in CZ range (300 m). It means that the influence of queuing is limited in one intersection rather than spreading to the upstream intersection, which is important for urban traffic control. In PCC+, the CAV is able to obtain a proper trajectory for itself. However, the HDVs around it stop the CAV to successfully undertake the optimized trajectory and the queuing for following HDVs behind the CAV is inevitable. What's worse, this might continue to jeopardize the optimization of the next CAV in upstream traffic flow. With HDV considered in the optimization and proper platoon chosen, the traffic mobility can be further improved as discussed in Section~\ref{Section:OptimalControlMode}. As shown in Fig.~\ref{fig:MC_700}, CAVs are able to lead the following HDVs to pass the intersection with proper velocity in MP.
Note that in MP algorithm, vehicles' idling behavior is avoided as much as possible, and therefore the fuel economy is significantly improved. The specific performance indexes are shown in Table~\ref{Table:Figure5Result}. Moreover, compared with PCC+ control in Fig.~\ref{fig:OC_700}, MP control in Fig.~\ref{fig:MC_700} creates much bigger gaps between trajectory blocks, which creates more optimization space for CAV. Generally speaking, the existence of these gaps means that the CAV algorithm brings lower traffic saturation and higher traffic mobility at the intersections~\cite{feng2018spatiotemporal}.
%It can be observed that in Fig.~\ref{fig:NC_700} the queuing at the signalized intersection is gradually accumulating, which means the intersection is over-saturated in $ 700 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $. On the contrary, with PCC+ and MP applied, the queuing accumulation is constrained in CZ range. It means queuing effect is limited in one intersection rather than spreading to the upstream intersection, which is important to urban traffic control.
%The reason is that the vehicles' stop-and-go behavior before the stopping line lowers the green phase utilization, which worsens the traffic efficiency. With CAV control applied to CAVs,
{
To further clarify the performance of MP algorithm in reducing fuel consumption, we present the trajectory of one single vehicle in Fig.~\ref{fig:ControlCompare:No}, indexed as no.~$73$ in Fig.~\ref{fig:ControlCompare}. Note that in No control of Fig.~\ref{fig:NC} or PCC+ control of Fig.~\ref{fig:OC}, the velocity remains $ 0 $ for a long time, which causes a huge amount of fuel consumption when the vehicle stops and the engine idles. The long-time acceleration and deceleration also make the engine less likely to work in high-efficiency working conditions, which worsens the fuel consumption performance. In MP control, on the contrary, the vehicle is always moving forward although the accel/decel cycles are increased moderately (but the oscillation amplitude has been significantly reduced) to arrive at the stopping line with a proper velocity at the proper time. The trajectory is much smoother and the idling behavior is avoided as much as possible. Consequently, the accumulative fuel consumption of this single vehicle under No control or PCC+ reaches $ 0.20\,\mathrm{L} $, while the MP control is only $ 0.12\,\mathrm{L} $.
}
% Consequently, the accumulative fuel consumption for No control or PCC+ reaches $ 0.2\,\mathrm{L} $, while the MP control is $ 1.2\,\mathrm{L} $. This result could apparently verify that the proposed method significant reduce the fuel consumption near the intersection.
\begin{table}[t]
\centering
\begin{tabular}[c]{cccc}
\toprule%[1.5pt]
& {\textbf{No Control}} & \textbf{PCC+} & {\textbf{Mixed Platoon}} \\
\midrule%[1pt]
{\textbf{ATTD} (s)} & $126.56$ & $125.26$ & $105.81$ \\
{\textbf{Fuel Consumption} ($ \mathrm{L}/100\mathrm{km} $)} & $16.77$ & $16.66$ & $10.53$ \\
{\textbf{Accel/Decel Cycles per Vehicle}} & $2.58$ & $8.64$ & $5.20$ \\
{\textbf{Ideling Time} (s)} & $40.29$ & $40.13$ & $2.32$ \\
\bottomrule%[1.5pt]
\end{tabular}
\caption{{Traffic efficiency and fuel consumption performance comparison of the three algorithm in Fig.~\ref{fig:ControlCompare}.}
\label{Table:Figure5Result}}
\end{table}
\begin{figure*}[t]
\centering
\subfigure[No control\label{fig:NC}] {\includegraphics[width=0.32\linewidth] {NoControl-eps-converted-to.pdf}}
\subfigure[PCC+ control\label{fig:OC}] {\includegraphics[width=0.32\linewidth] {PCCControl-eps-converted-to.pdf}}
\subfigure[Mixed platoon control\label{fig:MC}] {\includegraphics[width=0.32\linewidth] {MPControl-eps-converted-to.pdf}}
\caption{Trajectories of vehicle no.~$73$ under the three algorithms. Vehicle idling is avoided as much as possible in the MP algorithm compared to the benchmark algorithms. The smoothed velocity trajectory of the MP algorithm helps to improve fuel economy.\label{fig:ControlCompare:No}}
\end{figure*}
%Similar results have also been observed in.
%if the CAV control algorithm further takes the following HDVs into consideration, the CAV can initiatively guide the HDVs to pass the intersection with proper velocity. It can be seen from the evaluation index that MP algorithm has better performance than PCC+ in most of the cases.
%For traffic efficiency, the average travel time delay(ATTD) is chosen. For simplicity, the fuel consumption is transferred into fuel consumption per hundred kilometers.
\subsection{Simulation Results at Different Traffic Volumes}
\begin{figure*}[t]
\centering
\subfigure[Average Travel Time Delay Comparison] {\includegraphics[width=0.4\linewidth]{TrafficFlow_TravelTime-eps-converted-to.pdf}\label{}}
\hfil
\subfigure[Fuel Consumption Comparison] {\includegraphics[width=0.4\linewidth]{TrafficFlow_Fuel-eps-converted-to.pdf}\label{}}
\caption{{Performance comparison between the three algorithms in different traffic volumes (MPR = 50\%). Standard deviations of the data are also provided as error bars. The black lines means no optimized control input assigned to CAV. Blue lines and magenta lines are for PCC+ and MP algorithm.} \label{fig:ControlCompareSum}}
% In $ 600 \mathrm{-} 800 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $, PCC+ and mixed platoon perform almost the same. In $ 800 \mathrm{-} 1100 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$, mixed platoon algorithm outperforms PCC+ algorithm.
\end{figure*}
Fig.~\ref{fig:ControlCompareSum} compares the performance of the algorithms in various traffic volumes, which is also conducted in $ 50\% $ MPR of $ 100 $ vehicles. {Each working condition is simulated five times and standard deviations are plotted in the error bars as well. Generally speaking, MP has obvious improvement on both traffic efficiency and fuel consumption between $ 600 \mathrm{-} 1200 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $. When the traffic volume is lower than $ 600 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$, there is no queuing in the intersection no matter which control algorithm is applied to CAV. By contrast, in a much supersaturated traffic volume (\eg, over $ 1200 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$), the optimization space for CAV is rather low considering the high traffic density.
It is clearly observed that PCC+ shows limited improvement in traffic efficiency in the mixed traffic environment. Since HDVs are not considered in the optimization process in PCC+, more velocity fluctuations might happen, which makes the fuel consumption even worse compared with the case of no control applied to CAV.
%When the traffic volume is between $ 600 \mathrm{-} 800 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $, PCC+ and MP make similar improvement on traffic efficiency and fuel consumption. From $ 800 \mathrm{-} 1100 \, \mathrm{veh/(hour} \cdot \mathrm{lane}) $, the PCC+'s ATTD grows rapidly to almost the same degree as that of the case where no control is applied to CAV, and its fuel consumption exceeds that of the case without CAV control at around $ 1040 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$. However, MP still maintains lower ATTD and fuel consumption until around $ 1150 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$.
The reason why MP outperforms PCC+ is that when the traffic volume increases, the interference between HDVs and CAVs becomes much more frequent. When those HDVs are incorporated in CAV optimization like MP, the CAV can help improve the traffic efficiency and fuel consumption to a higher degree. }
% and the velocity fluctuation brings consumes more fuel than vehicles' idling at the stopping line
\subsection{Simulation Results at Different Traffic Volumes and MPRs}
\label{sec:FinalSimulation}
%\begin{table}[t]
% \centering
% \begin{tabular}[c]{ccccccccc}
% \toprule%[1.5pt]
% \multicolumn{2}{c}{} & \multicolumn{7}{c}{\textbf{Average Travel Time Delay (s)}} \\
% \midrule%[1pt]
% \multirow{2}{0.2\linewidth}{\centering\textbf{Traffic volume \\ $ \mathbf{veh/(hour \cdot lane)} $}} &
% \multirow{2}{*}{\centering\textbf{$ 600 $}} &
% \multirow{2}{*}{\centering\textbf{$ 700 $}} &
% \multirow{2}{*}{\centering\textbf{$ 800 $}} &
% \multirow{2}{*}{\centering\textbf{$ 900 $}} &
% \multirow{2}{*}{\centering\textbf{$ 1000 $}} &
% \multirow{2}{*}{\centering\textbf{$ 1100 $}} &
% \multirow{2}{*}{\centering\textbf{$ 1200 $}} \\
% & & & & & & & \\
% \midrule%[1pt]
% \multirow{2}{*}{MPR} $ 0\% $ & $ 92.54 $ & $ 114.86 $ & $ 132.10 $ & $ 151.22 $ & $ 163.67 $ & $ 178.98 $ & $ 178.92 $ \\
% $ 10\% $ & $ 92.54 $ & $ 114.86 $ & $ 132.10 $ & $ 151.22 $ & $ 163.67 $ & $ 178.98 $ & $ 178.92 $ \\
% & & & & & & & &\\
% \bottomrule%[1.5pt]
% \end{tabular}
% \caption{Simulation results comparison of $ 0\% $ and $ 100\% $ MPR under $ 600 \, \mathrm{and} \, 1000 \, \mathrm{veh/(hour \cdot lane)}$.}
% \label{tab:Result}
%\end{table}
% \multicolumn{2}{c}{\multirow{2}{*}{\begin{tabular}[c]{@{}c@{}} \textbf{Traffic volume}\\ $ \mathbf{veh/(hour \cdot lane)} $ \end{tabular}}} & \multirow{2}{*}{$ \mathbf{600} $} & \multirow{2}{*}{$ \mathbf{700} $} & \multirow{2}{*}{$ \mathbf{800} $} & \multirow{2}{*}{$ \mathbf{900} $} & \multirow{2}{*}{$ \mathbf{1000} $} & \multirow{2}{*}{$ \mathbf{1100} $} & \multirow{2}{*}{$ \mathbf{1200} $} \\
%\multicolumn{2}{c}{} & & & & & & & \\
In the last simulation, we verify MP algorithm performance in multiple MPRs and traffic volumes. In addition, a stochastic variant for the OVM model~\eqref{equ2:OVM_a} is also under consideration, which is given by
\begin{equation}
\dot{v}_{i}=\gamma \kappa[V_{\mathrm{des}}({d}_{i})-v_i],% \quad \mathrm{for $i \in [1,\mathrm{n}]$},
\end{equation}
where $ \gamma \in \left[0.9,1.1\right] $ denotes a stochastic sensitivity gain.
\begin{figure*}[t]
\centering
% \subfigure[Average Travel Time Delay of MP\label{fig:ATTD:raw}] {\includegraphics[width=0.45\linewidth]{DualPara_TravelTime_raw}}
% \hfil
% \subfigure[Fuel Consumption of MP\label{fig:FC:raw}] {\includegraphics[width=0.45\linewidth]{DualPara_Fuel_raw}}
\subfigure[ATTD Improvement\label{fig:ATTD:3D}] {\includegraphics[width=0.45\linewidth]{DualPara_TravelTime_Compare_3D-eps-converted-to.pdf}}
\hfil
\subfigure[Fuel Consumption Improvement\label{fig:FC:3D}] {\includegraphics[width=0.45\linewidth]{DualPara_Fuel_Compare_3D-eps-converted-to.pdf}}
\subfigure[ATTD Improvement in 2D\label{fig:ATTD:2D}] {\includegraphics[width=0.45\linewidth]{DualPara_TravelTime_Compare_2D-eps-converted-to.pdf}}
\hfil
\subfigure[Fuel Consumption Improvement in 2D\label{fig:FC:2D}] {\includegraphics[width=0.45\linewidth]{DualPara_Fuel_Compare_2D-eps-converted-to.pdf}}
\caption{Algorithm performance comparison in different traffic volumes and different MPRs. $ 0\% $ MPR is chosen as the reference value to evaluate the MP algorithm traffic benefits, which is shown as the red lines in Fig.~\ref{fig:ATTD:3D} $\sim$ Fig.~\ref{fig:FC:2D}.}
\end{figure*}
%{\color{red}When MPR is $ 0\% $, as the traffic volume increase, ATTD increases from $ 11.7 \, \mathrm{s} $ to $ 98.6 \, \mathrm{s} $ and fuel consumption increases from $ 2.22 \, \mathrm{L} $ to $ 6.55 \, \mathrm{L} $. In contrast, when MPR is $ 100\% $, ATTD increases from $ 11.3 \, \mathrm{s} $ to $ 87.1 \, \mathrm{s} $ and fuel consumption increases from $ 2.14 \, \mathrm{L} $ to $ 6.21 \, \mathrm{L} $. Generally speaking, ATTD is reduced by $ 3.42\% $-$ 11.7\% $ and fuel consumption is reduced by $ 3.15\% $-$ 5.19\% $ by MP.}
%
%{\color{blue} that the highest improvement in each traffic volume does not always occurs at the highest MPR.}
%, which means our algorithm may not find the best driving policy in high MPRs.
The specific results of ATTD and fuel consumption under MP are shown in Table~\ref{tab:Result:ATTD} and Table~\ref{tab:Result:FC} in~\ref{sec:Appendix_Simulation}. As can be observed, ATTD and fuel consumption get worse as traffic volume increase while the penetration of CAV helps improve the traffic mobility and fuel consumption. To present the result in a more straightforward way, the $ 0\% $ MPR is chosen as the reference value to evaluate the MP algorithm improvement, \emph{i.e.}, how much improvement MP algorithm brings compared to the case where all the vehicles are HDVs. As shown in Fig.\ref{fig:ATTD:3D} and~\ref{fig:FC:3D}, the red lines in the figure denote the reference plane of $ 0\% $ MPR and the surface means the MP algorithm improvements on the ATTD or fuel consumption. Similarly, the figures are re-plotted as 2D contours as shown in Fig.\ref{fig:ATTD:2D} and~\ref{fig:FC:2D}. It is shown that in general, the MP algorithm can improve the traffic efficiency and fuel consumption at the same time in $600\mathrm{-}1200 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$. The highest improvement occurs at around $ 35{-}40\% $ MPR in $1100\, \mathrm{veh/(hour} \cdot \mathrm{lane})$ and $ 80{-}85\% $ MPR in $1000\, \mathrm{veh/(hour} \cdot \mathrm{lane})$, with a $ 20\% $ improvement in ATTD and a $ 60\% $ improvement in fuel consumption.
We then make further investigations on the influence of traffic volumes. It is observed that the highest improvement of MP algorithm occurs at $1000\mathrm{-}1100$ $\mathrm{veh/(hour} \cdot \mathrm{lane})$, which meets our expectation on the algorithm. Given the maximum velocity $ v_{\max} $, it takes the first CAV $ L_{\mathrm{obs}}/v_{\max} $ time to arrive at CZ. At that time, the CAV is firstly optimized and the optimal following vehicle number $ n$ is obtained by~\eqref{equ2:T_Green:N}. Accordingly, in the best case, the number of the following vehicles perfectly matches our design, and the corresponding traffic density is $ L_{\mathrm{obs}}/(v_{\max} \cdot n) $. According to Table~\ref{tab3:OptimalControlMode} and Fig.~\ref{fig:DistanceNumber}, it can be obtained that the best traffic density is $ 3.3 \, \mathrm{s/veh} = 1080 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$ in our simulation scenario, which is consistent with our simulation results.
% When the MPR is quite low, the accumulative dynamics of a series of HDVs magnifies the fluctuation, and thus the disadvantages brought by the fluctuation exceed the advantages brought by CAV control.
When traffic volume reaches $1200 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$, the MP algorithm can hardly improve the traffic efficiency and fuel consumption, since in this high-density traffic condition, the queuing is inevitable and the optimization space for CAVs is rather small.
Finally, we analyze the influence of MPR. Note that existing research has pointed out that as MPR increases, the CAV's benefits on intersections may not increase in positive correlation; see, \eg,~\cite{ala2016modeling}.
%. For instance, Ala \etal~\cite{ala2016modeling} proposed Eco-CACC algorithm which includes queuing length into CAV control, where low MPR results show negative influence on fuel consumption because of HDV's unexpected behavior.
A similar result is also observed in our simulation. The proposed MP algorithm separates the mixed traffic flow into standard ``$ 1+n $" mixed platoons. The most crucial factor that influences the performance is whether all the HDVs are included in the mixed platoons. When MPR is relatively high (\eg, $ 70\% $), almost all the HDVs can be incorporated, which helps to greatly improve the traffic performance. However, when the MPR is extremely high (\eg, $ > 90\% $), the improvement percentage drops a little, which means that the MP algorithm is not the best policy for almost 100\% MPR scenarios. In general, our proposed algorithm shows evident positive improvement in almost all traffic conditions. The extensive simulation results confirm the great benefits of the MP algorithm in the mixed traffic intersection.
%It can be observed that when the MPR is higher than $ 40\% $, MP algorithm has positive influence on both traffic efficiency and fuel consumption in all traffic volumes. In the critical-saturated volume of around $700 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$, MP improves traffic efficiency by $ 60\% $ and fuel consumption by $ 25\% $. Even in the over-saturated volume of $900\mathrm{-}1000 \, \mathrm{veh/(hour} \cdot \mathrm{lane})$, it still has about $ 10\% $ improvement on the traffic efficiency and $ 5\% $ on the fuel consumption. In medium or high MPR, velocity fluctuation is limited in the overall optimal control of the mixed platoon, achieving the improvement of traffic efficiency and fuel consumption.
%setting up the proper terminal velocity and cooperative of vehicles
%In the two-lane scenario,
% We find that CAV control inevitably bring velocity fluctuation of HDVs. Long sequence of HDVs will amplify the velocity fluctuation like bullwhip effect, which worsens both traffic efficiency and fuel consumption.
\section{Conclusions}
\label{Section:Conclusion}
%Under system analysis, stability and controllability conditions of the proposed formation method are derived.
%It can be applied in any MPRs of mixed traffic environment, which proves the universality of the proposed mixed platoon formation.
In this paper, the notion of ``$ 1+n $" mixed platoon for CAV control in mixed traffic intersections has been proposed. Assigned as the leading vehicle of $ n $ HDVs, the potential of the CAV in improving the global traffic mobility at the intersection has been revealed. Based on rigorous theoretical analysis, we have pointed out that the proposed mixed platoon formation is open-loop stable and controllable in a very mild condition, which is irrelevant to the mixed platoon size $ n $.
An optimal control framework has been established with consideration of velocity deviation and fuel consumption of the whole mixed platoon, where the terminal velocity has also been optimized to improve traffic throughput. Furthermore, a hierarchical event-triggered algorithm has been designed to solve the collision problem between adjacent mixed platoons, which can be applied in any MPR of mixed traffic environments. Traffic simulations have verified the effectiveness of the proposed optimal control method.
%The main conclusions are as follows:
%\begin{enumerate}[(1)]
% \item The mixed platoon' stability and controllability conditions are only related to HDVs' car-following model. The proposed mixed platoon formation is stable and controllable in a very mild condition, which is also irrelevant to the mixed platoon size $ n $.
% \item Considering constant traffic SPAT scenario, most signal CAV algorithms set terminal velocity as the highest velocity. It is proved that differ from single CAV, the terminal velocity of a mixed platoon should be adapted to an optimized value rather than the highest velocity to maximize the traffic mobility.
% \item In simulation, we verify that the proposed mixed platoon algorithm outperforms single CAV algorithm in different traffic volumes and find that CAV algorithm may not have satisfying result in rather low MPRs.
%\end{enumerate}
%It is proved that the mixed platoon algorithm is able to extend the ability of single CAV algorithm by initiatively considering HDVs in the optimization.
{One future direction is to address the problem of possible heterogeneous dynamics and model uncertainties in HDVs' behaviors in the optimal control framework for the ``$1+n$'' mixed platoon. %These issues have been recently discussed in car-following scenarios through controllability analysis~\cite{wang2020controllability} and model estimation~\cite{jin2018connected}.
Considering that multiple CAVs could exist in a mixed platoon when passing the intersection in the same green phasing time, especially in a high-MPR scenario, another interesting topic is to apply cooperative control algorithms to CAVs to further improve the overall performance for the mixed traffic intersection. Moreover, note that this paper focuses on the longitudinal control of CAVs by forbidding lane changing in CZ, addressing the lane changing behavior formally for both CAVs and HDVs is also a significant future direction.} Finally, field experiments are also needed for further validation of the proposed algorithm.
%The further research are still needed in several aspects. Firstly, speed fluctuation of HDVs can be studied by using head-to-tail method. Secondly, HDVs' accidentally cut-in behavior has not been considered in mixed platoon algorithm. Finally, {\color{red}Heterogeneous case~\cite{wang2020controllability}.} And the reinforcement learning method can also be introduced in to mixed traffic environment studies to overcome the model uncertainty problem.
\section*{Acknowledgments}
This work was supported by the National Key Research and Development Program of China under Grant\\ 2018YFE0204302, the National Natural Science Foundation of China under Grant 52072212, the Key-Area Research and Development Program of Guangdong Province under Grant 2019B090912001, Institute of China Intelligent and Connect Vehicles (CICV) and Tsinghua University-Didi Joint Research Center for Future Mobility.
\appendix
\section{Proof of Theorem~\ref{Th:Stability}}
\label{sec:Appendix}
We consider the ``$1+n$'' mixed platoon stability consists of $ 1 $ leading CAV and $ n $ following HDVs. Firstly, we consider the initial case of $ n=1 $, \emph{i.e.}, there is only one HDV following the leading CAV.
To obtain the eigenvalue $\lambda$ of $A_{n = 1}$, we need to solve $ \left| \lambda I - A_{n = 1} \right| = 0 $, leading to
$
{\lambda}^{4} + {\alpha_{2}}{\lambda}^{3} + {\alpha_{1}} {\lambda}^{2} = 0,
$
whose solutions are
\begin{equation}
\label{equ2:n2:EigenValue}
\begin{gathered}
\lambda_{1} =0,\; \lambda_{2} =0,\\
\lambda_{3} =-\frac{\alpha_{2}}{2} - \frac{\sqrt{\alpha_{2}^{2} - 4 \alpha_{1}}}{2},\;
\lambda_{4} =-\frac{\alpha_{2}}{2} + \frac{\sqrt{\alpha_{2}^{2} - 4 \alpha_{1}}}{2}.
\end{gathered}
\end{equation}
The two zero eigenvalues indicate that the system is not asymptotically stable, but can be made critically stable. According to Definition~\ref{def:Stability}, we have the stability criterion as $\alpha_{1}>0, \alpha_{2}>0$.
Then, we assume the system is stable when $ n=k $, \ie, all the eigenvalues of $A_{n=k}$ have negative real parts, and focus on the case where $ n=k+1 $. The system matrix can be expressed as
\begin{equation}
A_{n = k + 1}=\left[
\begin{array}{ccc}
{{A}_{n = k}} & {0} & {0}\\
{1} & {0} & {-1}\\
{\alpha_{3}} & {\alpha_{1}} & {-\alpha_{2}}
\end{array}\right].
\end{equation}
Given a block matrix
$$
P = \left[ \begin{array}{cc}
A & B \\
C & D
\end{array}\right],
$$
if $ D $ is invertible, it holds that\cite{greub2012linear}
$$
|P| = \left| \begin{array}{cc}
A & B \\
C & D
\end{array}\right| = \left|A-BD^{-1}C\right|\left|D\right| .
$$
According to the initial state assumption, we have that
$$
\begin{bmatrix}
{\lambda} & {1}\\
{-\alpha_{1}} & {\lambda + \alpha_{2}}
\end{bmatrix}
$$
is invertible, and thus it is obtained that
\begin{equation}
\left| \lambda I - {A}_{n = k + 1} \right|
= \left|\lambda I - {A}_{n = k}\right|\left|\begin{array}{cc}
{\lambda} & {1}\\
{-\alpha_{1}} & {\lambda + \alpha_{2}}
\end{array}\right|
=\left|\lambda I - {A}_{n = k}\right| (\lambda^2+
\alpha_2 \lambda+\alpha_1).
\end{equation}
According to the assumption, we only need to consider the solutions of $ \lambda^2+
\alpha_2 \lambda+\alpha_1=0$, which are stable if and only if $\alpha_{1}>0, \alpha_{2}>0$. Hence, we have that ${A}_{n = k + 1} $ is stable if and only if $\alpha_{1}>0, \alpha_{2}>0$, which completes the proof of Theorem~\ref{Th:Stability} according to the method of mathematical induction.
\section{Performance Indexes of Simulations under Different Traffic Volumes and MPRs}
\label{sec:Appendix_Simulation}
In this section, we present the specific data of performance indexes under MP control at different traffic volumes and MPRs in Section~\ref{sec:FinalSimulation}.
\begin{table}[h]
\centering
\begin{tabular}{cccccccc}
\toprule
\diagbox[width=12em]{\textbf{MPR}}{\textbf{Traffic volume} \\ $ \mathbf{(vphpl)} $} & $ \mathbf{600} $ & $ \mathbf{700} $ & $ \mathbf{800} $ & $ \mathbf{900} $ & $ \mathbf{1000} $ & $ \mathbf{1100} $ & $ \mathbf{1200} $ \\
\midrule
{$ \mathbf{0\%} $} & $ 92.54 $ & $ 114.86 $ & $ 132.10 $ & $ 151.22 $ & $ 163.67 $ & $ 178.98 $ & $ 178.92 $ \\
\cline{2-8}
{$ \mathbf{20\%} $} & $ 92.43 $ & $ 98.51 $ & $ 130.26 $ & $ 137.99 $ & $ 153.73 $ & $ 162.17 $ & $ 169.13 $ \\
{$ \mathbf{40\%} $} & $ 92.31 $ & $ 93.12 $ & $ 128.53 $ & $ 139.12 $ & $ 154.69 $ & $ 175.66 $ & $ 156.51 $ \\
{$ \mathbf{60\%} $} & $ 92.11 $ & $ 104.69 $ & $ 122.42 $ & $ 144.64 $ & $ 155.61 $ & $ 167.25 $ & $ 168.55 $ \\
{$ \mathbf{80\%} $} & $ 92.27 $ & $ 101.41 $ & $ 100.20 $ & $ 122.91 $ & $ 148.48 $ & $ 158.63 $ & $ 165.96 $ \\
{$ \mathbf{100\%} $} & $ 92.27 $ & $ 97.75 $ & $ 107.10 $ & $ 136.39 $ & $ 158.79 $ & $ 156.97 $ & $ 159.95 $ \\
\bottomrule
\end{tabular}
\caption{Average travel time delay (s) simulation results of $ 0\% $ and $ 100\% $ MPR under $ 600 \, \mathrm{to} \, 1200 \, \mathrm{veh/(hour \cdot lane)}$ traffic volume.}
\label{tab:Result:ATTD}
\end{table}
\begin{table}[h]
\centering
\begin{tabular}{cccccccc}
\toprule
\diagbox[width=12em]{\textbf{MPR}}{\textbf{Traffic volume} \\ $ \mathbf{(vphpl)} $} & $ \mathbf{600} $ & $ \mathbf{700} $ & $ \mathbf{800} $ & $ \mathbf{900} $ & $ \mathbf{1000} $ & $ \mathbf{1100} $ & $ \mathbf{1200} $ \\
\midrule
{$ \mathbf{0\%} $} & $ 3.81 $ & $ 12.2 $ & $ 21.66 $ & $ 35.29 $ & $ 45.68 $ & $ 59.64 $ & $ 58.23 $ \\
\cline{2-8}
{$ \mathbf{20\%} $} & $ 3.78 $ & $ 6.28 $ & $ 19.17 $ & $ 23.57 $ & $ 34.37 $ & $ 40.31 $ & $ 48.12 $ \\
{$ \mathbf{40\%} $} & $ 4.06 $ & $ 4.18 $ & $ 17.59 $ & $ 24.94 $ & $ 35.25 $ & $ 51.31 $ & $ 37.42 $ \\
{$ \mathbf{60\%} $} & $ 3.76 $ & $ 8.45 $ & $ 16.19 $ & $ 28.80 $ & $ 36.78 $ & $ 46.57 $ & $ 42.07 $ \\
{$ \mathbf{80\%} $} & $ 4.04 $ & $ 7.20 $ & $ 6.47 $ & $ 15.61 $ & $ 33.20 $ & $ 36.47 $ & $ 44.87 $ \\
{$ \mathbf{100\%} $} & $ 4.08 $ & $ 5.79 $ & $ 9.07 $ & $ 23.32 $ & $ 37.32 $ & $ 38.60 $ & $ 39.12 $ \\
\bottomrule
\end{tabular}
\caption{Fuel consumption ($ \mathrm{L}/100\mathrm{km} $) simulation results of $ 0\% $ and $ 100\% $ MPR under $ 600 \, \mathrm{to} \, 1200 \, \mathrm{veh/(hour \cdot lane)}$ traffic volume.}
\label{tab:Result:FC}
\end{table}
%\section*{References}
\bibliography{mybibfile}
% \bibliographystyle{elsarticle-num}
\end{document}
\begin{thebibliography}{47}
\expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
\providecommand{\url}[1]{\texttt{#1}}
\providecommand{\href}[2]{#2}
\providecommand{\path}[1]{#1}
\providecommand{\DOIprefix}{doi:}
\providecommand{\ArXivprefix}{arXiv:}
\providecommand{\URLprefix}{URL: }
\providecommand{\Pubmedprefix}{pmid:}
\providecommand{\doi}[1]{\href{http://dx.doi.org/#1}{\path{#1}}}
\providecommand{\Pubmed}[1]{\href{pmid:#1}{\path{#1}}}
\providecommand{\bibinfo}[2]{#2}
\ifx\xfnm\relax \def\xfnm[#1]{\unskip,\space#1}\fi
%Type = Article
\bibitem[{Akcelik(1989)}]{akcelik1989efficiency}
\bibinfo{author}{Akcelik, R.} (\bibinfo{year}{1989}).
\newblock \bibinfo{title}{Efficiency and drag in the power-based model of fuel
consumption}.
\newblock {\it \bibinfo{journal}{Transportation Research Part B:
Methodological}\/}, {\it \bibinfo{volume}{23}\/}, \bibinfo{pages}{376--385}.
%Type = Article
\bibitem[{Ala et~al.(2016)Ala, Yang \& Rakha}]{ala2016modeling}
\bibinfo{author}{Ala, M.~V.}, \bibinfo{author}{Yang, H.}, \&
\bibinfo{author}{Rakha, H.} (\bibinfo{year}{2016}).
\newblock \bibinfo{title}{Modeling evaluation of eco--cooperative adaptive
cruise control in vicinity of signalized intersections}.
\newblock {\it \bibinfo{journal}{Transportation Research Record}\/}, {\it
\bibinfo{volume}{2559}\/}, \bibinfo{pages}{108--119}.
%Type = Article
\bibitem[{Asadi \& Vahidi(2011)}]{asadi2011predictive}
\bibinfo{author}{Asadi, B.}, \& \bibinfo{author}{Vahidi, A.}
(\bibinfo{year}{2011}).
\newblock \bibinfo{title}{Predictive cruise control: Utilizing upcoming traffic
signal information for improving fuel economy and reducing trip time}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Control Systems
Technology}\/}, {\it \bibinfo{volume}{19}\/}, \bibinfo{pages}{707--714}.
%Type = Article
\bibitem[{Bian et~al.(2019)Bian, Li, Ren, Wang, Li \&
Liu}]{bian2019cooperation}
\bibinfo{author}{Bian, Y.}, \bibinfo{author}{Li, S.~E.}, \bibinfo{author}{Ren,
W.}, \bibinfo{author}{Wang, J.}, \bibinfo{author}{Li, K.}, \&
\bibinfo{author}{Liu, H.} (\bibinfo{year}{2019}).
\newblock \bibinfo{title}{Cooperation of multiple connected vehicles at
unsignalized intersections: Distributed observation, optimization, and
control}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Industrial
Electronics}\/}, {\it \bibinfo{volume}{67}\/},
\bibinfo{pages}{10744--10754}.
%Type = Article
\bibitem[{Contreras-Castillo et~al.(2017)Contreras-Castillo, Zeadally \&
Guerrero-Iba{\~n}ez}]{contreras2017internet}
\bibinfo{author}{Contreras-Castillo, J.}, \bibinfo{author}{Zeadally, S.}, \&
\bibinfo{author}{Guerrero-Iba{\~n}ez, J.~A.} (\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Internet of vehicles: architecture, protocols, and
security}.
\newblock {\it \bibinfo{journal}{IEEE Internet of Things Journal}\/}, {\it
\bibinfo{volume}{5}\/}, \bibinfo{pages}{3701--3709}.
%Type = Inproceedings
\bibitem[{Cui et~al.(2017)Cui, Seibold, Stern \& Work}]{cui2017stabilizing}
\bibinfo{author}{Cui, S.}, \bibinfo{author}{Seibold, B.},
\bibinfo{author}{Stern, R.}, \& \bibinfo{author}{Work, D.~B.}
(\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Stabilizing traffic flow via a single autonomous
vehicle: Possibilities and limitations}.
\newblock In {\it \bibinfo{booktitle}{2017 IEEE Intelligent Vehicles Symposium
(IV)}\/} (pp. \bibinfo{pages}{1336--1341}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Inproceedings
\bibitem[{Du et~al.(2017)Du, HomChaudhuri \& Pisu}]{du2017coordination}
\bibinfo{author}{Du, Z.}, \bibinfo{author}{HomChaudhuri, B.}, \&
\bibinfo{author}{Pisu, P.} (\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Coordination strategy for vehicles passing multiple
signalized intersections: A connected vehicle penetration rate study}.
\newblock In {\it \bibinfo{booktitle}{2017 American control conference
(ACC)}\/} (pp. \bibinfo{pages}{4952--4957}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Article
\bibitem[{Elnagar et~al.(1995)Elnagar, Kazemi \&
Razzaghi}]{elnagar1995pseudospectral}
\bibinfo{author}{Elnagar, G.}, \bibinfo{author}{Kazemi, M.~A.}, \&
\bibinfo{author}{Razzaghi, M.} (\bibinfo{year}{1995}).
\newblock \bibinfo{title}{The pseudospectral {Legendre} method for discretizing
optimal control problems}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Automatic Control}\/},
{\it \bibinfo{volume}{40}\/}, \bibinfo{pages}{1793--1796}.
%Type = Article
\bibitem[{Fang et~al.(2020)Fang, Yang, Wang \& Jing}]{fang2020trajectory}
\bibinfo{author}{Fang, S.}, \bibinfo{author}{Yang, L.}, \bibinfo{author}{Wang,
T.}, \& \bibinfo{author}{Jing, S.} (\bibinfo{year}{2020}).
\newblock \bibinfo{title}{Trajectory planning method for mixed vehicles
considering traffic stability and fuel consumption at the signalized
intersection}.
\newblock {\it \bibinfo{journal}{Journal of Advanced Transportation}\/}, {\it
\bibinfo{volume}{2020}\/}.
%Type = Article
\bibitem[{Feng et~al.(2015)Feng, Head, Khoshmagham \&
Zamanipour}]{feng2015real}
\bibinfo{author}{Feng, Y.}, \bibinfo{author}{Head, K.~L.},
\bibinfo{author}{Khoshmagham, S.}, \& \bibinfo{author}{Zamanipour, M.}
(\bibinfo{year}{2015}).
\newblock \bibinfo{title}{A real-time adaptive signal control in a connected
vehicle environment}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{55}\/}, \bibinfo{pages}{460--473}.
%Type = Article
\bibitem[{Feng et~al.(2018)Feng, Yu \& Liu}]{feng2018spatiotemporal}
\bibinfo{author}{Feng, Y.}, \bibinfo{author}{Yu, C.}, \& \bibinfo{author}{Liu,
H.~X.} (\bibinfo{year}{2018}).
\newblock \bibinfo{title}{Spatiotemporal intersection control in a connected
and automated vehicle environment}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{89}\/}, \bibinfo{pages}{364--383}.
%Type = Inproceedings
\bibitem[{Gerla et~al.(2014)Gerla, Lee, Pau \& Lee}]{gerla2014internet}
\bibinfo{author}{Gerla, M.}, \bibinfo{author}{Lee, E.-K.},
\bibinfo{author}{Pau, G.}, \& \bibinfo{author}{Lee, U.}
(\bibinfo{year}{2014}).
\newblock \bibinfo{title}{Internet of vehicles: From intelligent grid to
autonomous cars and vehicular clouds}.
\newblock In {\it \bibinfo{booktitle}{2014 IEEE world forum on internet of
things (WF-IoT)}\/} (pp. \bibinfo{pages}{241--246}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Article
\bibitem[{Gipps(1981)}]{gipps1981behavioural}
\bibinfo{author}{Gipps, P.~G.} (\bibinfo{year}{1981}).
\newblock \bibinfo{title}{A behavioural car-following model for computer
simulation}.
\newblock {\it \bibinfo{journal}{Transportation Research Part B:
Methodological}\/}, {\it \bibinfo{volume}{15}\/}, \bibinfo{pages}{105--111}.
%Type = Book
\bibitem[{Greub(2012)}]{greub2012linear}
\bibinfo{author}{Greub, W.~H.} (\bibinfo{year}{2012}).
\newblock {\it \bibinfo{title}{Linear algebra}\/} volume~\bibinfo{volume}{23}.
\newblock \bibinfo{publisher}{Springer Science \& Business Media}.
%Type = Article
\bibitem[{Helbing \& Tilch(1998)}]{helbing1998generalized}
\bibinfo{author}{Helbing, D.}, \& \bibinfo{author}{Tilch, B.}
(\bibinfo{year}{1998}).
\newblock \bibinfo{title}{Generalized force model of traffic dynamics}.
\newblock {\it \bibinfo{journal}{Physical Review E}\/}, {\it
\bibinfo{volume}{58}\/}, \bibinfo{pages}{133}.
%Type = Article
\bibitem[{Jiang et~al.(2017)Jiang, Hu, An, Wang \& Park}]{jiang2017eco}
\bibinfo{author}{Jiang, H.}, \bibinfo{author}{Hu, J.}, \bibinfo{author}{An,
S.}, \bibinfo{author}{Wang, M.}, \& \bibinfo{author}{Park, B.~B.}
(\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Eco approaching at an isolated signalized
intersection under partially connected and automated vehicles environment}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{79}\/}, \bibinfo{pages}{290--307}.
%Type = Article
\bibitem[{Jin \& Orosz(2014)}]{jin2014dynamics}
\bibinfo{author}{Jin, I.~G.}, \& \bibinfo{author}{Orosz, G.}
(\bibinfo{year}{2014}).
\newblock \bibinfo{title}{Dynamics of connected vehicle systems with delayed
acceleration feedback}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{46}\/}, \bibinfo{pages}{46--64}.
%Type = Article
\bibitem[{Lee \& Park(2012)}]{lee2012development}
\bibinfo{author}{Lee, J.}, \& \bibinfo{author}{Park, B.}
(\bibinfo{year}{2012}).
\newblock \bibinfo{title}{Development and evaluation of a cooperative vehicle
intersection control algorithm under the connected vehicles environment}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Intelligent
Transportation Systems}\/}, {\it \bibinfo{volume}{13}\/},
\bibinfo{pages}{81--90}.
%Type = Article
\bibitem[{Li et~al.(2015)Li, Xu, Huang, Cheng \& Peng}]{li2015eco}
\bibinfo{author}{Li, S.~E.}, \bibinfo{author}{Xu, S.}, \bibinfo{author}{Huang,
X.}, \bibinfo{author}{Cheng, B.}, \& \bibinfo{author}{Peng, H.}
(\bibinfo{year}{2015}).
\newblock \bibinfo{title}{Eco-departure of connected vehicles with {V2X}
communication at signalized intersections}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Vehicular Technology}\/},
{\it \bibinfo{volume}{64}\/}, \bibinfo{pages}{5439--5449}.
%Type = Article
\bibitem[{Liang et~al.(2019)Liang, Guler \& Gayah}]{liang2019joint}
\bibinfo{author}{Liang, X.}, \bibinfo{author}{Guler, S.~I.}, \&
\bibinfo{author}{Gayah, V.~V.} (\bibinfo{year}{2019}).
\newblock \bibinfo{title}{Joint optimization of signal phasing and timing and
vehicle speed guidance in a connected and autonomous vehicle environment}.
\newblock {\it \bibinfo{journal}{Transportation Research Record}\/}, {\it
\bibinfo{volume}{2673}\/}, \bibinfo{pages}{70--83}.
%Type = Article
\bibitem[{Lighthill \& Whitham(1955)}]{lighthill1955kinematic}
\bibinfo{author}{Lighthill, M.~J.}, \& \bibinfo{author}{Whitham, G.~B.}
(\bibinfo{year}{1955}).
\newblock \bibinfo{title}{On kinematic waves {II}. a theory of traffic flow on
long crowded roads}.
\newblock {\it \bibinfo{journal}{Proceedings of the Royal Society of London.
Series A. Mathematical and Physical Sciences}\/}, {\it
\bibinfo{volume}{229}\/}, \bibinfo{pages}{317--345}.
%Type = Inproceedings
\bibitem[{Lopez et~al.(2018)Lopez, Behrisch, Bieker-Walz, Erdmann,
Fl{\"o}tter{\"o}d, Hilbrich, L{\"u}cken, Rummel, Wagner \&
WieBner}]{lopez2018microscopic}
\bibinfo{author}{Lopez, P.~A.}, \bibinfo{author}{Behrisch, M.},
\bibinfo{author}{Bieker-Walz, L.}, \bibinfo{author}{Erdmann, J.},
\bibinfo{author}{Fl{\"o}tter{\"o}d, Y.-P.}, \bibinfo{author}{Hilbrich, R.},
\bibinfo{author}{L{\"u}cken, L.}, \bibinfo{author}{Rummel, J.},
\bibinfo{author}{Wagner, P.}, \& \bibinfo{author}{WieBner, E.}
(\bibinfo{year}{2018}).
\newblock \bibinfo{title}{Microscopic traffic simulation using sumo}.
\newblock In {\it \bibinfo{booktitle}{2018 21st International Conference on
Intelligent Transportation Systems (ITSC)}\/} (pp.
\bibinfo{pages}{2575--2582}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Article
\bibitem[{Ma et~al.(2017)Ma, Li, Zhou, Hu \& Park}]{ma2017parsimonious}
\bibinfo{author}{Ma, J.}, \bibinfo{author}{Li, X.}, \bibinfo{author}{Zhou, F.},
\bibinfo{author}{Hu, J.}, \& \bibinfo{author}{Park, B.~B.}
(\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Parsimonious shooting heuristic for trajectory design
of connected automated traffic part {II}: computational issues and
optimization}.
\newblock {\it \bibinfo{journal}{Transportation Research Part B:
Methodological}\/}, {\it \bibinfo{volume}{95}\/}, \bibinfo{pages}{421--441}.
%Type = Book
\bibitem[{Maciejowski(2002)}]{maciejowski2002predictive}
\bibinfo{author}{Maciejowski, J.~M.} (\bibinfo{year}{2002}).
\newblock {\it \bibinfo{title}{Predictive control: with constraints}\/}.
\newblock \bibinfo{publisher}{Pearson education}.
%Type = Article
\bibitem[{Malikopoulos et~al.(2018)Malikopoulos, Cassandras \&
Zhang}]{malikopoulos2018decentralized}
\bibinfo{author}{Malikopoulos, A.~A.}, \bibinfo{author}{Cassandras, C.~G.}, \&
\bibinfo{author}{Zhang, Y.~J.} (\bibinfo{year}{2018}).
\newblock \bibinfo{title}{A decentralized energy-optimal control framework for
connected automated vehicles at signal-free intersections}.
\newblock {\it \bibinfo{journal}{Automatica}\/}, {\it \bibinfo{volume}{93}\/},
\bibinfo{pages}{244--256}.
%Type = Article
\bibitem[{Milan{\'e}s et~al.(2010)Milan{\'e}s, P{\'e}rez, Onieva \&
Gonz{\'a}lez}]{Milanes10}
\bibinfo{author}{Milan{\'e}s, V.}, \bibinfo{author}{P{\'e}rez, J.},
\bibinfo{author}{Onieva, E.}, \& \bibinfo{author}{Gonz{\'a}lez, C.}
(\bibinfo{year}{2010}).
\newblock \bibinfo{title}{Controller for urban intersections based on wireless
communications and fuzzy logic}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Intelligent
Transportation Systems}\/}, {\it \bibinfo{volume}{11}\/},
\bibinfo{pages}{243--248}.
%Type = Article
\bibitem[{Onieva et~al.(2015)Onieva, Hern{\'a}ndez-Jayo, Osaba, Perallos \&
Zhang}]{onieva2015multi}
\bibinfo{author}{Onieva, E.}, \bibinfo{author}{Hern{\'a}ndez-Jayo, U.},
\bibinfo{author}{Osaba, E.}, \bibinfo{author}{Perallos, A.}, \&
\bibinfo{author}{Zhang, X.} (\bibinfo{year}{2015}).
\newblock \bibinfo{title}{A multi-objective evolutionary algorithm for the
tuning of fuzzy rule bases for uncoordinated intersections in autonomous
driving}.
\newblock {\it \bibinfo{journal}{Information Sciences}\/}, {\it
\bibinfo{volume}{321}\/}, \bibinfo{pages}{14--30}.
%Type = Article
\bibitem[{Patterson \& Rao(2014)}]{patterson2014gpops}
\bibinfo{author}{Patterson, M.~A.}, \& \bibinfo{author}{Rao, A.~V.}
(\bibinfo{year}{2014}).
\newblock \bibinfo{title}{Gpops-{II}: A {MATLAB} software for solving
multiple-phase optimal control problems using hp-adaptive {Gaussian}
quadrature collocation methods and sparse nonlinear programming}.
\newblock {\it \bibinfo{journal}{ACM Transactions on Mathematical Software
(TOMS)}\/}, {\it \bibinfo{volume}{41}\/}, \bibinfo{pages}{1}.
%Type = Inproceedings
\bibitem[{Priemer \& Friedrich(2009)}]{Priemer09}
\bibinfo{author}{Priemer, C.}, \& \bibinfo{author}{Friedrich, B.}
(\bibinfo{year}{2009}).
\newblock \bibinfo{title}{A decentralized adaptive traffic signal control using
{V2I} communication data}.
\newblock In {\it \bibinfo{booktitle}{2009 12th International IEEE Conference
on Intelligent Transportation Systems}\/} (pp. \bibinfo{pages}{1--6}).
\newblock \bibinfo{address}{St. Louis, MO, USA}: \bibinfo{organization}{IEEE}.
%Type = Article
\bibitem[{Rakha et~al.(2001)Rakha, Kang \& Dion}]{rakha2001estimating}
\bibinfo{author}{Rakha, H.}, \bibinfo{author}{Kang, Y.-S.}, \&
\bibinfo{author}{Dion, F.} (\bibinfo{year}{2001}).
\newblock \bibinfo{title}{Estimating vehicle stops at undersaturated and
oversaturated fixed-time signalized intersections}.
\newblock {\it \bibinfo{journal}{Transportation Research Record}\/}, {\it
\bibinfo{volume}{1776}\/}, \bibinfo{pages}{128--137}.
%Type = Inproceedings
\bibitem[{Salman et~al.(2017)Salman, Ozdemir \& Celebi}]{Salman17}
\bibinfo{author}{Salman, M.~A.}, \bibinfo{author}{Ozdemir, S.}, \&
\bibinfo{author}{Celebi, F.~V.} (\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Fuzzy logic based traffic surveillance system using
cooperative {V2X} protocols with low penetration rate}.
\newblock In {\it \bibinfo{booktitle}{2017 International Symposium on Networks,
Computers and Communications (ISNCC)}\/} (pp. \bibinfo{pages}{1--6}).
\newblock \bibinfo{address}{Marrakech, Morocco}: \bibinfo{organization}{IEEE}.
%Type = Book
\bibitem[{Skogestad \& Postlethwaite(2007)}]{skogestad2007multivariable}
\bibinfo{author}{Skogestad, S.}, \& \bibinfo{author}{Postlethwaite, I.}
(\bibinfo{year}{2007}).
\newblock {\it \bibinfo{title}{Multivariable feedback control: analysis and
design}\/} volume~\bibinfo{volume}{2}.
\newblock \bibinfo{publisher}{Wiley New York}.
%Type = Article
\bibitem[{Stern et~al.(2018)Stern, Cui, Delle~Monache, Bhadani
et~al.}]{stern2018dissipation}
\bibinfo{author}{Stern, R.~E.}, \bibinfo{author}{Cui, S.},
\bibinfo{author}{Delle~Monache, M.~L.}, \bibinfo{author}{Bhadani, R.} et~al.
(\bibinfo{year}{2018}).
\newblock \bibinfo{title}{Dissipation of stop-and-go waves via control of
autonomous vehicles: Field experiments}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{89}\/}, \bibinfo{pages}{205--221}.
%Type = Article
\bibitem[{Treiber \& Kesting(2013)}]{treiber2013traffic}
\bibinfo{author}{Treiber, M.}, \& \bibinfo{author}{Kesting, A.}
(\bibinfo{year}{2013}).
\newblock \bibinfo{title}{Traffic flow dynamics}.
\newblock {\it \bibinfo{journal}{Traffic Flow Dynamics: Data, Models and
Simulation, Springer-Verlag Berlin Heidelberg}\/}, .
%Type = Inproceedings
\bibitem[{Vinitsky et~al.(2018)Vinitsky, Parvate, Kreidieh, Wu \&
Bayen}]{vinitsky2018lagrangian}
\bibinfo{author}{Vinitsky, E.}, \bibinfo{author}{Parvate, K.},
\bibinfo{author}{Kreidieh, A.}, \bibinfo{author}{Wu, C.}, \&
\bibinfo{author}{Bayen, A.} (\bibinfo{year}{2018}).
\newblock \bibinfo{title}{Lagrangian control through deep-{RL}: Applications to
bottleneck decongestion}.
\newblock In {\it \bibinfo{booktitle}{2018 21st International Conference on
Intelligent Transportation Systems (ITSC)}\/} (pp.
\bibinfo{pages}{759--765}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Inproceedings
\bibitem[{Wang et~al.(2020{\natexlab{a}})Wang, Zheng, Chen, Xu \&
Li}]{wang2020leading}
\bibinfo{author}{Wang, J.}, \bibinfo{author}{Zheng, Y.}, \bibinfo{author}{Chen,
C.}, \bibinfo{author}{Xu, Q.}, \& \bibinfo{author}{Li, K.}
(\bibinfo{year}{2020}{\natexlab{a}}).
\newblock \bibinfo{title}{Leading cruise control in mixed traffic flow}.
\newblock In {\it \bibinfo{booktitle}{Conference on Decision and Control
(CDC)}\/} (pp. \bibinfo{pages}{1--7}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Article
\bibitem[{Wang et~al.(2020{\natexlab{b}})Wang, Zheng, Chen, Xu \&
Li}]{wang2020leadingSCS}
\bibinfo{author}{Wang, J.}, \bibinfo{author}{Zheng, Y.}, \bibinfo{author}{Chen,
C.}, \bibinfo{author}{Xu, Q.}, \& \bibinfo{author}{Li, K.}
(\bibinfo{year}{2020}{\natexlab{b}}).
\newblock \bibinfo{title}{Leading cruise control in mixed traffic flow: System
modeling, controllability, and string stability}.
\newblock {\it \bibinfo{journal}{arXiv preprint arXiv:2012.04313}\/}, .
%Type = Article
\bibitem[{Wang et~al.(2020{\natexlab{c}})Wang, Zheng, Xu, Wang \&
Li}]{wang2020controllability}
\bibinfo{author}{Wang, J.}, \bibinfo{author}{Zheng, Y.}, \bibinfo{author}{Xu,
Q.}, \bibinfo{author}{Wang, J.}, \& \bibinfo{author}{Li, K.}
(\bibinfo{year}{2020}{\natexlab{c}}).
\newblock \bibinfo{title}{Controllability analysis and optimal control of mixed
traffic flow with human-driven and autonomous vehicles}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Intelligent
Transportation Systems}\/}, (pp. \bibinfo{pages}{1--15}).
%Type = Article
\bibitem[{Wilson \& Ward(2011)}]{wilson2011car}
\bibinfo{author}{Wilson, R.~E.}, \& \bibinfo{author}{Ward, J.~A.}
(\bibinfo{year}{2011}).
\newblock \bibinfo{title}{Car-following models: fifty years of linear stability
analysis--a mathematical perspective}.
\newblock {\it \bibinfo{journal}{Transportation Planning and Technology}\/},
{\it \bibinfo{volume}{34}\/}, \bibinfo{pages}{3--18}.
%Type = Inproceedings
\bibitem[{Wu et~al.(2018)Wu, Bayen \& Mehta}]{wu2018stabilizing}
\bibinfo{author}{Wu, C.}, \bibinfo{author}{Bayen, A.~M.}, \&
\bibinfo{author}{Mehta, A.} (\bibinfo{year}{2018}).
\newblock \bibinfo{title}{Stabilizing traffic with autonomous vehicles}.
\newblock In {\it \bibinfo{booktitle}{2018 IEEE International Conference on
Robotics and Automation (ICRA)}\/} (pp. \bibinfo{pages}{1--7}).
\newblock \bibinfo{organization}{IEEE}.
%Type = Inproceedings
\bibitem[{Wu et~al.(2017)Wu, Kreidieh, Vinitsky \& Bayen}]{wu2017emergent}
\bibinfo{author}{Wu, C.}, \bibinfo{author}{Kreidieh, A.},
\bibinfo{author}{Vinitsky, E.}, \& \bibinfo{author}{Bayen, A.~M.}
(\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Emergent behaviors in mixed-autonomy traffic}.
\newblock In {\it \bibinfo{booktitle}{Conference on Robot Learning}\/} (pp.
\bibinfo{pages}{398--407}).
%Type = Article
\bibitem[{Xu et~al.(2018{\natexlab{a}})Xu, Ban, Bian, Li, Wang, Li \&
Li}]{xu2018cooperative}
\bibinfo{author}{Xu, B.}, \bibinfo{author}{Ban, X.~J.}, \bibinfo{author}{Bian,
Y.}, \bibinfo{author}{Li, W.}, \bibinfo{author}{Wang, J.},
\bibinfo{author}{Li, S.~E.}, \& \bibinfo{author}{Li, K.}
(\bibinfo{year}{2018}{\natexlab{a}}).
\newblock \bibinfo{title}{Cooperative method of traffic signal optimization and
speed control of connected vehicles at isolated intersections}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Intelligent
Transportation Systems}\/}, {\it \bibinfo{volume}{20}\/},
\bibinfo{pages}{1390--1403}.
%Type = Article
\bibitem[{Xu et~al.(2018{\natexlab{b}})Xu, Li, Bian, Li, Ban, Wang \&
Li}]{xu2018distributed}
\bibinfo{author}{Xu, B.}, \bibinfo{author}{Li, S.~E.}, \bibinfo{author}{Bian,
Y.}, \bibinfo{author}{Li, S.}, \bibinfo{author}{Ban, X.~J.},
\bibinfo{author}{Wang, J.}, \& \bibinfo{author}{Li, K.}
(\bibinfo{year}{2018}{\natexlab{b}}).
\newblock \bibinfo{title}{Distributed conflict-free cooperation for multiple
connected vehicles at unsignalized intersections}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{93}\/}, \bibinfo{pages}{322--334}.
%Type = Article
\bibitem[{Yang et~al.(2017)Yang, Rakha \& Ala}]{yang2017eco}
\bibinfo{author}{Yang, H.}, \bibinfo{author}{Rakha, H.}, \&
\bibinfo{author}{Ala, M.~V.} (\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Eco-cooperative adaptive cruise control at signalized
intersections considering queue effects}.
\newblock {\it \bibinfo{journal}{IEEE Transactions on Intelligent
Transportation Systems}\/}, {\it \bibinfo{volume}{18}\/},
\bibinfo{pages}{1575--1585}.
%Type = Article
\bibitem[{Zhao et~al.(2018)Zhao, Ngoduy, Shepherd, Liu \&
Papageorgiou}]{zhao2018platoon}
\bibinfo{author}{Zhao, W.}, \bibinfo{author}{Ngoduy, D.},
\bibinfo{author}{Shepherd, S.}, \bibinfo{author}{Liu, R.}, \&
\bibinfo{author}{Papageorgiou, M.} (\bibinfo{year}{2018}).
\newblock \bibinfo{title}{A platoon based cooperative eco-driving model for
mixed automated and human-driven vehicles at a signalised intersection}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{95}\/}, \bibinfo{pages}{802--821}.
%Type = Article
\bibitem[{Zheng \& Liu(2017)}]{zheng2017estimating}
\bibinfo{author}{Zheng, J.}, \& \bibinfo{author}{Liu, H.~X.}
(\bibinfo{year}{2017}).
\newblock \bibinfo{title}{Estimating traffic volumes for signalized
intersections using connected vehicle data}.
\newblock {\it \bibinfo{journal}{Transportation Research Part C: Emerging
Technologies}\/}, {\it \bibinfo{volume}{79}\/}, \bibinfo{pages}{347--362}.
%Type = Article
\bibitem[{Zheng et~al.(2020)Zheng, Wang \& Li}]{zheng2020smoothing}
\bibinfo{author}{Zheng, Y.}, \bibinfo{author}{Wang, J.}, \&
\bibinfo{author}{Li, K.} (\bibinfo{year}{2020}).
\newblock \bibinfo{title}{Smoothing traffic flow via control of autonomous
vehicles}.
\newblock {\it \bibinfo{journal}{IEEE Internet of Things Journal}\/}, {\it
\bibinfo{volume}{7}\/}, \bibinfo{pages}{3882--3896}.
\end{thebibliography}