A Saturation-Based Optimal Velocity Model for Traffic Flow Dynamics
Extracted source text · arXiv 2509.22671
\documentclass[12pt]{elsarticle}
\usepackage{amsmath, amssymb, amsthm}
\usepackage{geometry, graphicx, multicol, float}
\geometry{margin=1in}
\usepackage{url}
%\newtheorem{proposition}{Proposition}
\begin{document}
\begin{frontmatter}
\title{A Hybrid Optimal Velocity–Drag Model for Traffic Flow Dynamics}
\author[umich]{Nizhum Rahman}
\author[umich]{Trachette L. Jackson}
\address[umich]{Department of Mathematics, University of Michigan, Ann Arbor, MI 48109, United States}
\begin{abstract}
We develop a \emph{Hybrid Optimal Velocity--Drag} (OVD) model that combines the behavioral
structure of the classical \emph{Optimal Velocity Model} (OVM) with a
drag-based saturation law motivated by Newtonian mechanics. The model
retains the OVM principle that desired speed depends on headway, but
replaces the linear relaxation law with a formulation that enforces
bounded accelerations and smooth convergence to equilibrium. After
reviewing the OVM framework and its stability properties, we introduce
the OVD dynamics and derive the dispersion relation from a
linear stability analysis of uniform flow. This analysis shows that the
OVD formulation preserves the instability mechanisms responsible for
stop-and-go waves, while avoiding the unrealistic acceleration
predictions of the classical OVM. We also illustrate the model using
desired velocity functions of hyperbolic tangent type, a common choice
in the OVM literature, and highlight applications to large-scale traffic
simulation and empirical data studies. Taken together, our results
demonstrate that the OVD model enforces bounded accelerations while
preserving the nonlinear instabilities of the classical OVM, providing
a physically grounded and interpretable foundation for advancing
traffic flow research.
\end{abstract}
\begin{keyword}
traffic flow modeling \sep optimal velocity model \sep hybrid optimal velocity--drag model \sep linear stability analysis \sep stop-and-go waves
\end{keyword}
%\noindent\textbf{Keywords:} traffic flow modeling; optimal velocity model; hybrid OVM--Drag; linear stability analysis; stop-and-go waves
\end{frontmatter}
\section{Introduction}
Traffic flow is a classic example of how complex patterns can emerge from simple interactions: phenomena like waves and traffic jams arise from the interactions of many individual vehicles. At the microscopic level, these interactions are often described by \emph{car-following models}, which explain how drivers adjust their speed based on the distance to other vehicles and the relative speed of those vehicles.
One of the most influential of these models is the Optimal Velocity Model (OVM), introduced in the 1990s by Bando et al.\cite{Bando1995}. Despite being simple, the OVM can reproduce key traffic phenomena, including stop-and-go waves, the formation of traffic jams, and other unstable (nonlinear) patterns, making it a key reference in the study of traffic flow within physics, engineering, and applied mathematics\cite{Helbing2001,Nagatani2002,Kerner2004,Wilson2011}.
Over the past three decades, many extensions and variations of the OVM have been developed to better capture real-world traffic behavior. For example, Jiang et al.\cite{Jiang2001} added a term to account for differences in speed between successive vehicles (leading to the \emph{Full Velocity Difference Model}, FVDM). Treiber et al.\cite{Treiber2000} introduced the idea that drivers anticipate the movement of multiple cars ahead, not just the one directly in front. Other developments include adding random fluctuations to represent differences among drivers~\cite{Chowdhury2000}, extending the model to multiple lanes~\cite{Herman1997}, and including drivers’ reaction time delays~\cite{Zhang1999}. Researchers have also experimented with different forms of the optimal velocity function itself. The original OVM used a hyperbolic tangent (tanh) function for its optimal velocity, which does not allow vehicles to fully stop; later studies proposed modified functions (using cutoffs or piecewise definitions) that let the model realistically simulate cars coming to a complete stop~\cite{Li2012,Ge2004}.
At the same time, the OVM has been widely investigated from the perspective of nonlinear dynamics. It has been shown to produce traveling waves in traffic, solitary wave patterns (solitons), and metastable states — flows that appear steady but can break down under small disturbances~\cite{Komatsu1995}. Researchers use tools like linear stability analysis and dispersion relations to determine when a steady flow of traffic becomes unstable, and experiments such as Sugiyama et al.’s circular-track study~\cite{Sugiyama2008} have confirmed that traffic jams can form spontaneously, even without any obvious bottleneck. These findings demonstrate that OVM is not only a practical tool for traffic modeling but also a rich system for studying nonlinear traffic phenomena~\cite{Orosz2010,Treiber2013}.
However, the OVM and most of its variants have a fundamental limitation: they assume a linear response rule, meaning a vehicle’s acceleration is proportional to the difference between its current speed and its desired speed. While this assumption is convenient for calculations, it does not reflect how real vehicles behave. In reality, a car’s acceleration is determined by a balance between the engine’s power and forces like aerodynamic drag. Classical mechanics shows that under constant engine power and a drag force that increases with the square of speed (quadratic drag), a car’s speed will smoothly approach a certain maximum value — often modeled with a hyperbolic tangent function~\cite{Batchelor1967}. This kind of drag-based speed saturation is well known in fluid mechanics, and it appears in scenarios ranging from falling objects to projectiles and vehicles encountering air resistance. Yet, these physical considerations are rarely included in OVM-type models. As a result, most car-following models still use the simple linear acceleration assumption, leaving a gap between the behavior-based descriptions of driver response and a more physically realistic representation of vehicle dynamics~\cite{Wilson2008,Seibold2013}. From a practical standpoint, having a limit on acceleration is also essential for ensuring driver comfort, improving fuel efficiency, and enabling effective automated vehicle control~\cite{Milanes2014,Kesting2010}.
To address this issue, we propose a Hybrid Optimal Velocity--Drag (OVD) model.
This model keeps the core idea of OVM — in which a driver’s desired speed depends on the headway (the distance to the car in front) —
but it replaces the linear acceleration rule with a quadratic saturation rule inspired by drag force physics.
This modification ensures that speeding up and slowing down are smooth and physically consistent,
while still preserving the OVM’s ability to produce traffic waves and other nonlinear patterns.
We begin by reviewing the classical OVM.
Then, we introduce the OVD model and analyze its stability using the dispersion relation.
Finally, we present numerical simulations to illustrate the model’s dynamics and discuss their implications for traffic flow modeling and large-scale simulation~\cite{Punzo2016}.
\section{Classical Optimal Velocity Model}
The Optimal Velocity Model (OVM), introduced by Bando et al.~\cite{Bando1995}, is one of the most widely studied microscopic car-following models. In this framework, each driver adjusts their acceleration based on the difference between their current speed and a desired speed, which depends on the distance (headway) to the car in front.
Suppose we have $N$ vehicles moving on a ring road of length $L$. We denote the position and velocity of the $n$-th vehicle at time $t$ by $x_n(t)$ and $v_n(t)$, respectively. The headway, that is, the spacing between two successive vehicles, is given by
\[
\Delta x_n(t) \;=\; x_{n+1}(t) - x_n(t),
\]
with periodic indexing $x_{N+1} = x_1 + L$ to close the ring.
In the OVM, the equations of motion are
\begin{equation}
\label{eq:ovm}
\frac{dx_n}{dt} = v_n,
\qquad
\frac{dv_n}{dt} = \alpha \big(V(\Delta x_n) - v_n\big),
\end{equation}
where $\alpha > 0$ is a sensitivity parameter. The function $V(\Delta x)$, known as the \emph{optimal velocity function}, denotes the driver’s desired speed for a given headway. In this way, the model captures the basic behavioral rule that drivers speed up when the gap in front increases, and slow down when it decreases.
\subsection{Interpretation}
The OVM describes the simple rule that drivers speed up when they are moving slower than the desired speed $V(\Delta x_n)$ and slow down when they are moving faster. The sensitivity parameter $\alpha$ determines how strongly and quickly this adjustment happens. The function $V(\Delta x)$ is taken to be monotonically increasing, ranging from zero up to a maximum speed, and represents the safe-driving response to the space available ahead. At this stage, we do not specify an exact formula for $V(\Delta x)$; only its general properties are important for developing more advanced models. In later sections, we will introduce concrete forms of $V(\Delta x)$ as needed for analysis and numerical examples.
\section{Hybrid OVD model}
The Optimal Velocity Model (OVM) captures the idea that drivers adjust their acceleration according to the difference between their actual velocity and a headway-dependent target velocity. While elegant, the OVM uses a \emph{linear relaxation law}, which means that the larger the difference $V(\Delta x_n)-v_n$, the stronger the acceleration. This can yield unrealistically large accelerations and does not directly reflect vehicle dynamics.
\subsection{Physical motivation from drag-limited motion}
In physics, a single vehicle subject to a constant forward force from the engine and a quadratic aerodynamic drag force satisfies
\begin{equation}
\label{eq:drag}
m \dot v = F_{\text{eng}} - k v^2,
\end{equation}
where $m$ is the vehicle’s mass, $F_{\text{eng}}$ is the effective forward force generated by the engine,
and $k v^2$ represents aerodynamic drag, which grows proportionally to the square of the velocity.
Here $k$ is a drag constant that depends on air density, frontal area, and the drag coefficient.
Defining
\[
a := \frac{F_{\text{eng}}}{m},
\qquad
v_{\max} := \sqrt{\frac{F_{\text{eng}}}{k}},
\]
we obtain the rearranged form
\begin{equation}
\dot v = a\left(1 - \frac{v^2}{v_{\max}^2}\right).
\end{equation}
This law has two appealing properties:
\begin{itemize}
\item At low speeds, $\dot v \approx a$, so the vehicle initially accelerates with constant rate $a$.
\item As $v\to v_{\max}$, the acceleration smoothly decays to zero, so the vehicle saturates to a finite speed instead of accelerating without bound.
\end{itemize}
Solving this ODE with initial condition $v(0)=v_0$ gives
\[
v(t) = v_{\max}\,\tanh\!\left(\tfrac{a}{v_{\max}}t + \operatorname{artanh}\!\tfrac{v_0}{v_{\max}}\right),
\]
which explicitly shows that the velocity approaches $v_{\max}$ as $t\to\infty$.
Thus, drag-limited dynamics naturally lead to a \emph{hyperbolic tangent convergence} of velocity, which is both physically grounded and smooth.
\subsection{Formulation of the hybrid model}
In traffic, the relevant ``terminal velocity'' is not a fixed $v_{\max}$ but rather the desired speed determined by the spacing. To combine the behavioral realism of the OVM with drag-based saturation, we replace $v_{\max}$ in the drag law by the instantaneous desired speed $V(\Delta x_n)$. The resulting Hybrid Optimal Velocity--Drag (OVD) model is
\begin{equation}
\label{eq:hybrid}
\frac{dx_n}{dt} = v_n,
\qquad
\frac{dv_n}{dt} = a\left(1 - \frac{v_n^2}{V(\Delta x_n)^2}\right),
\end{equation}
where $a>0$ is an acceleration scale and $V(\Delta x_n)$ is a monotonically increasing, bounded function of headway (not specified explicitly here).
Note that $a$ is constant, while the effective sensitivity
$\alpha_{\mathrm{eff}}(v,V)$ depends on the current state.
\subsection{Interpretation of the dynamics}
Equation~\eqref{eq:hybrid}, which consists of the position update and the velocity dynamics, can be interpreted in three regimes:
\begin{itemize}
\item \textbf{Small headways:} When $\Delta x_n$ is small, the desired speed $V(\Delta x_n)$ is close to zero, so the velocity equation in~\eqref{eq:hybrid} is negative. This leads to smooth deceleration and avoids collisions.
\item \textbf{Large headways:} When $\Delta x_n$ is large, $V(\Delta x_n)\approx v_{\max}$ (the free-flow speed), and the velocity equation reduces to
\[
\dot v = a\!\left(1-\frac{v^2}{v_{\max}^2}\right),
\]
i.e., the standard drag-limited law, giving smooth convergence to the free-flow speed.
\item \textbf{Intermediate headways:} For moderate gaps, vehicles accelerate when $v_n < V(\Delta x_n)$, but as $v_n$ approaches $V(\Delta x_n)$, the factor $1-(v_n/V)^2$ dampens further acceleration, producing gradual convergence.
\end{itemize}
\subsection{Relation to the Classical OVM}
Rewriting the hybrid velocity law,
\[
\dot v = a\left(1-\frac{v^2}{V^2}\right)
= \frac{a(V+v)}{V^2}(V-v),
\]
shows that the structure is the same as in the OVM, except that the sensitivity
is no longer constant but depends on the state:
\[
\alpha_{\text{eff}}(v,V) = \frac{a(V+v)}{V^2}.
\]
\subsection{Local correspondence with the OVM}
Although the Hybrid Optimal Velocity--Drag (OVD) model is nonlinear in general,
it reduces to the classical OVM structure in the neighborhood of an equilibrium
velocity. Let $v = V + \delta v$ with $|\delta v| \ll V$. Substituting
into the hybrid law gives
\[
\dot v = \frac{a(V+v)}{V^2}(V-v)
= -\frac{a}{V^2}(2V+\delta v)\,\delta v.
\]
Expanding to leading order in $\delta v$ yields
\[
\dot v \approx \frac{2a}{V}\,(V - v).
\]
Thus, near equilibrium, the hybrid formulation coincides with the OVM
form but with an effective sensitivity
\[
\alpha_{\text{eff}} = \frac{2a}{V}.
\]
This calculation shows that while the Hybrid OVD model enforces
bounded acceleration and nonlinear saturation away from equilibrium,
its local behavior resembles the OVM. In particular, small deviations
from equilibrium are governed by a linear relaxation law with an effective
sensitivity depending on both the drag parameter $a$ and the equilibrium
velocity $V$.
\begin{figure}[H]
\centering
\includegraphics[width=0.9\textwidth]{OVM_vs_Hybrid.pdf}
\caption{Comparison between the classical OVM and the Hybrid OVD
model. \textbf{(a)} Acceleration laws $\dot v$ versus velocity $v$
for fixed $V=20$ m/s. \textbf{(b)} Effective sensitivity. The Hybrid
model enforces bounded acceleration and a velocity-dependent
effective sensitivity, whereas the classical OVM uses a linear law
with constant sensitivity.}
\label{fig:ovm-hybrid-comparison}
\end{figure}
This comparison highlights the physical interpretation of the Hybrid
OVD model. While the classical OVM assumes a linear response with
constant sensitivity, the hybrid formulation ensures that acceleration
remains bounded and decreases smoothly as vehicles approach their target
speed. In other words, drivers accelerate strongly when far from
equilibrium but naturally ease off as they get closer, reflecting the
influence of aerodynamic drag and engine limitations. The Hybrid model
therefore preserves the familiar local structure of the OVM near
equilibrium while extending it with more realistic global behavior.
Detailed derivations and plots of the corresponding \emph{jerk},
$j=d^2v/dt^2$, are provided in~\ref{app:jerk-analysis}.
\subsection{Limitations}
\label{sec:limitations}
As shown in~\ref{app:jerk-analysis}, the Hybrid OVD model
produces a cubic dependence of jerk on velocity that is well behaved for
$0 \leq v \leq V$, vanishing at both $v=0$ and $v=V$. Beyond the desired
velocity, however, the jerk grows cubically with $v$. In typical highway
traffic such overspeeding is rare, so the effect is not critical for most
applications. Still, it highlights an inherent limitation of the model.
A natural refinement would be to replace the quadratic saturation term
with a saturating nonlinear mapping, such as a $\tanh$- or $\arctan$-based
law, which would keep the jerk uniformly bounded across all velocities
while preserving the correct local dynamics. Developing and testing such
variants remains a promising direction for future work, particularly in
settings involving automated or aggressive driving where vehicles may
temporarily exceed $V$.
\section{Stability Analysis of the Hybrid OVD Model}
\subsection{Model and steady (uniform-flow) solution}
We consider
\begin{equation}
\label{eq:hyb}
\frac{dx_n}{dt}=v_n,
\qquad
\frac{dv_n}{dt}=a\!\left(1-\frac{v_n^2}{V(\Delta x_n)^2}\right),
\qquad
\Delta x_n := x_{n+1}-x_n,
\end{equation}
on a ring of length $L$ with $N$ vehicles ($x_{n+N}=x_n+L$).
Seek a uniformly translating equilibrium
\[
x_n^0(t)=nb+ct,\qquad v_n^0=c,
\]
with constant headway $b=L/N$ so that $\Delta x_n^0=b$ for all $n$.
Setting $\dot v_n^0=0$ in \eqref{eq:hyb} yields
\[
0=a\!\left(1-\frac{c^2}{V(b)^2}\right)\;\Rightarrow\; c=\pm V(b).
\]
Taking forward motion gives the uniform flow
\begin{equation}
\label{eq:uniform}
x_n^0(t)=nb+V(b)\,t,\qquad v_n^0=V(b),\qquad b=L/N.
\end{equation}
\subsection{Linearization with position and velocity perturbations}
Introduce small perturbations
\[
x_n=x_n^0+\xi_n,\qquad v_n=v_n^0+\eta_n,
\]
so the perturbed headway is $\Delta x_n=b+(\xi_{n+1}-\xi_n)$.
Let $c:=V(b)$ and $f:=V'(b)$. Linearizing \eqref{eq:hyb} about $(\Delta x_n,v_n)=(b,c)$ gives
\begin{align}
\dot\xi_n &= \eta_n, \label{eq:lin1}\\
\dot\eta_n &= -\frac{2a}{c}\,\eta_n \;+\; \frac{2a}{c}\,f\,(\xi_{n+1}-\xi_n).
\label{eq:lin2}
\end{align}
\subsection{Normal modes and dispersion relation}
Seek normal modes
\[
\xi_n=\hat\xi\,e^{\lambda t+i k n},\qquad
\eta_n=\hat\eta\,e^{\lambda t+i k n},
\quad k=\frac{2\pi m}{N},\; m=0,\dots,N-1.
\]
From \eqref{eq:lin1}–\eqref{eq:lin2}:
\[
\lambda \hat\xi=\hat\eta,\qquad
\lambda \hat\eta=-\frac{2a}{c}\,\hat\eta+\frac{2af}{c}\,(e^{ik}-1)\hat\xi.
\]
Eliminating $\hat\eta=\lambda\hat\xi$ yields the quadratic eigenvalue equation
\begin{equation}
\boxed{\;
\lambda^2 \;+\; \frac{2a}{c}\,\lambda \;+\; \frac{2af}{c}\,\big(1-e^{ik}\big)=0,
\;}
\label{eq:dispersion-hybrid}
\end{equation}
the \emph{dispersion relation} of the hybrid model.
\subsection{Long-wave expansion and stability threshold}
For small $k$ (long waves), $e^{ik}=1+ik-\tfrac{1}{2}k^2+O(k^3)$.
With the ansatz $\lambda(k)=i\mu k+\nu k^2+O(k^3)$ and matching powers, one finds $\mu=f$ and
\[
\nu=\frac{c}{2a}\,f^2-\frac{f}{2}.
\]
Thus
\[
\lambda(k)=i f k+\Big(\frac{c}{2a}f^2-\frac{f}{2}\Big)k^2+O(k^3).
\]
The leading real part is
\[
\Re \lambda \;\approx\; \Big(\frac{c}{2a}f^2-\frac{f}{2}\Big)k^2,
\]
which is negative provided
\[
\boxed{\;a>a^\ast,\qquad a^\ast:=c\,V'(b).\;}
\]
Hence uniform flow is linearly stable for $a>a^\ast$, unstable for $a<a^\ast$, and neutrally stable at $a=a^\ast$.
The imaginary part gives $\omega=\Im\lambda\approx f k$, showing that long waves propagate at speed $V'(b)$ in the comoving frame.
\subsection{Eigenvalue spectra and complex-plane illustration}
To validate this threshold, we compute the full eigenvalue spectra for the illustrative case $V(s)=\tanh(s)$, where $c=\tanh(b)$ and $f=\operatorname{sech}^2(b)$. The dispersion relation \eqref{eq:dispersion-hybrid} then yields two eigenvalues
\[
\lambda_\pm(k)=
-\frac{a}{c}\;\pm\;\frac{1}{2}\sqrt{\left(\frac{2a}{c}\right)^2
-4\,\frac{2af}{c}\bigl(1-e^{ik}\bigr)},
\]
for each mode $k=2\pi m/N$. We compare spectra above and below the threshold:
\[
\text{Stable: } a=2.2\,a^\ast,\qquad
\text{Unstable: } a=0.5\,a^\ast.
\]
According to standard stability theory, an equilibrium is stable if all eigenvalues lie in the open left half-plane (\(\Re(\lambda)<0\)) and unstable if any eigenvalue enters the right half-plane \cite{Strogatz2015,Hirsch2012,Khalil2002}. This spectral test is widely used in dynamical systems, control theory, and even the analysis of biological travelling waves~\cite{Rahman2022}.
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{hybrid_eigs_complexplane.pdf}
\caption{Eigenvalue spectra of the Hybrid OVD model with \(V(s)=\tanh(s)\).
\textbf{Left:} Stable regime (\(a > a^\ast\)): all eigenvalues lie in the left half-plane, so perturbations decay.
\textbf{Right:} Unstable regime (\(a < a^\ast\)): several eigenvalues cross into the right half-plane, leading to exponential growth of disturbances and stop-and-go waves.}
\label{fig:hybrid-eigs-complex}
\end{figure}
\section{Numerical Illustration}
We now provide numerical simulations of the Hybrid OVD model on a
ring road to compare with the classical results reported in \cite{Bando1995}.
The setup follows the same protocol: a periodic system of
$N=100$ vehicles is simulated using forward Euler integration with step
$\Delta t=0.1$ over a time horizon $T=100$. Two representative ring
lengths are chosen:
\[
L_{\text{stable}}=200, \qquad L_{\text{unstable}}=50,
\]
corresponding to equilibrium headways $b=L/N=2.0$ and $b=0.5$,
respectively. For each case, the acceleration parameter $a$ is set above
or below the theoretical threshold $a^\ast=c\,V'(b)$, so that one
simulation is linearly stable while the other is unstable.
\subsection{Trajectory and velocity plots}
Initial conditions consist of uniform spacing $b=L/N$ with a small
perturbation added to the first car position. The resulting unwrapped
trajectories $x_n(t)$ and velocities $v_n(t)$ are recorded for all cars.
As in the original study, we display a subset of evenly spaced vehicles
to illustrate the collective behavior.
Figure~\ref{fig:hybrid-trajectories-allcars} shows that in the stable case
the trajectories remain nearly parallel and the perturbation decays, whereas
in the unstable case nonlinear stop-and-go waves emerge.
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{hybrid_trajectories_allcars.pdf}
\caption{Unwrapped trajectories of all $N=100$ vehicles.
\textbf{Left:} stable regime ($L=200$, $b=2.0$, $a/a^\ast=2.20$),
perturbations decay and trajectories remain nearly parallel.
\textbf{Right:} unstable regime ($L=50$, $b=0.5$, $a/a^\ast=0.50$),
perturbations amplify into nonlinear stop-and-go waves.}
\label{fig:hybrid-trajectories-allcars}
\end{figure}
For clearer observation of the traffic jam dynamics, we include in~\ref{app:longtraj} an additional unstable simulation extended
to $t=300$, where the stop-and-go pattern is more pronounced.
In Fig.~\ref{fig:velocities_10cars}, we report the velocity evolution $v_n(t)$
for a subset of representative cars under both stable and unstable regimes.
In the stable case, the velocities remain close to equilibrium, exhibiting
only small oscillations that gradually decay over time.
At $t=0$, all cars start from a velocity near the equilibrium value
$v_n(0) \approx 0.97$, and this remains nearly uniform across the fleet.
By contrast, in the unstable case, although the initial velocity is also
uniform across cars, it starts at a lower value $v_n(0) \approx 0.46$.
The perturbation then amplifies with time, leading to the formation of
pronounced stop-and-go waves. Notably, the velocities never reach zero;
vehicles slow down considerably but do not come to a complete stop,
which is consistent with the smooth form of the optimal velocity function
\(
V(\Delta x) = \tanh(\Delta x)
\).
These results confirm that this choice, with the same parameters as before,
correctly reproduces the bifurcation between stable and unstable traffic flow dynamics.
It is worth emphasizing, however, that the $\tanh$--based optimal velocity
function prevents vehicles from ever reaching a full stop. This feature,
already noted in the original formulation by Bando et al.~\cite{Bando1995},
may limit the model’s ability to capture realistic jammed states in which
cars come to rest. As highlighted in subsequent reviews~\cite{Helbing2001,Nagatani2002},
several modified OVM-type models have introduced sharper or cutoff forms of
$V(\Delta x)$ to address this limitation, allowing cars to stop under dense
traffic conditions~\cite{Jiang2001,Li2012}.
\begin{figure}[H]
\centering
\includegraphics[width=0.99\textwidth]{hybrid_velocities_10cars.pdf}
\caption{Velocity evolution $v_n(t)$ for selected cars in both stable
(left) and unstable (right) regimes. In the stable case, cars start at
$v_n(0) \approx 0.97$ (normalized by $v_{\max}$) and remain close to uniform flow.
In the unstable case, cars start at $v_n(0) \approx 0.46$ and develop
stop-and-go oscillations as the perturbation grows. The oscillations
correspond to significant slowdowns, but velocities never reach zero.}
\label{fig:velocities_10cars}
\end{figure}
\subsection{Fourier mode analysis}
To better understand how perturbations evolve, we analyze the Fourier
amplitudes of the headway deviations. Defining
\[
y_n(t) = x_n(t) - \big(nb + c\,t\big),
\]
we compute
\[
A_k(t) = \left| \sum_{n=0}^{N-1} y_n(t)\,e^{-i \alpha_k n} \right|,
\qquad \alpha_k=\frac{2\pi k}{N}.
\]
Following the original study, we focus on five representative modes,
\[
k=10,\;20,\;30,\;40,\;50.
\]
The behavior of these amplitudes directly indicates stability. In the
stable regime, all $A_k(t)$ decrease steadily over time, reflecting the
fact that perturbations die out when $a>a^\ast$. In the unstable regime,
one or more modes grow instead, showing the amplification of
disturbances predicted by the dispersion relation.
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{hybrid_fourier_modes.pdf}
\caption{Fourier amplitudes $A_k(t)$ for modes $k=10,20,30,40,50$ in the
Hybrid OVD model with $N=100$ vehicles.
\textbf{Left:} stable regime ($L=200$, $b=2.0$, $a/a^\ast=2.20$), where all
Fourier modes decay over time.
\textbf{Right:} unstable regime ($L=50$, $b=0.5$, $a/a^\ast=0.50$), where
selected modes grow, indicating the formation of stop-and-go waves in line
with the dispersion relation.}
\label{fig:hybrid-fourier}
\end{figure}
\section{Implications and Applications}
The Hybrid OVD model developed here combines two complementary
perspectives on car-following: the behavioral realism of the Optimal
Velocity Model, where desired speeds depend on headway, and the physical
saturation dynamics of drag-based acceleration. By replacing the linear
relaxation term of the OVM with a nonlinear drag law, the hybrid model
ensures bounded accelerations and smooth convergence to equilibrium
speeds. This modification eliminates the unrealistically large
accelerations and oscillations that can occur in the classical
formulation, while preserving the essential feature that driver response
is fundamentally governed by available headway.
From a modeling perspective, the parameter $a$ provides an interpretable
acceleration scale linked to driver aggressiveness or vehicle power,
while the headway response $V(\Delta x)$ can be specified to match
empirical observations. The model is therefore well suited for
calibration against microscopic trajectory datasets, whether from
controlled experiments or naturalistic driving studies. In particular,
datasets such as NGSIM have been widely used for calibration and
validation of car-following models~\cite{Punzo2016}, and established
methodologies are available in the literature~\cite{Treiber2013}. At
the same time, the nonlinear saturation law improves the robustness of
the model for large-scale simulations, where bounded velocities and
stable dynamics are critical.
\subsection{Connection to prior OVM studies}
Although the theoretical framework is general in $V(\Delta x)$, a widely
used specification in the OVM literature is the sigmoid map
\[
V(\Delta x) = \frac{v_{\max}}{2}\left[\tanh\!\left(\frac{\Delta x-h_c}{\ell}\right)+1\right],
\]
with parameters $v_{\max}$ (free-flow speed), $h_c$ (critical headway),
and $\ell$ (slope parameter). This form, or its logistic equivalent, has
been employed extensively in classical OVM studies
\cite{Helbing2001,Treiber2013}, offering both analytical
tractability and a realistic saturation toward free-flow speed.
In this study we do not analyze this specific functional form in detail;
our illustrative results instead use the simpler hyperbolic tangent
mapping for clarity. Nevertheless, within the Hybrid OVD framework,
adopting the sigmoid law would not change the structure of the model but
would enable direct comparison with prior work. The parameters $h_c$ and
$\ell$ determine the inflection point and steepness of the headway–speed
relation, and thus directly affect the linear stability threshold. This
makes the model suitable for calibration with real data while retaining
the physically motivated saturation dynamics introduced by the drag law.
In this sense, the Hybrid OVD model can be viewed as a natural extension
of the classical OVM: the same widely used $V(\Delta x)$ function is
preserved, but the velocity evolution law is enriched with a physically
grounded, drag-inspired nonlinearity.
\subsection{Applications}
The Hybrid OVD model opens several practical directions. First, it can
serve as a core component in multi-lane and network-level simulations,
where bounded velocity laws are essential for numerical stability.
Second, the explicit role of the acceleration parameter $a$ offers a
straightforward way to represent heterogeneity across vehicles and
drivers: higher values capture aggressive or powerful vehicles, while
lower values describe cautious or underpowered ones. Third, the model is
directly relevant to mixed traffic composed of both human-driven and
automated vehicles, where calibration of $V(\Delta x)$ and physically
consistent acceleration dynamics are both critical. In this context,
realistic car-following behavior is central to the design of adaptive
cruise control systems~\cite{Milanes2014}, and has been shown to affect
traffic capacity in automated driving strategies~\cite{Kesting2010}.
Finally, the hybrid law may inform control-oriented applications such as
ACC, where balancing responsiveness with comfort requires realistic but
stable acceleration behavior.
In summary, the Hybrid OVD model provides a bridge between behavioral
and physical approaches to car-following. By enriching the classical OVM
with a drag-based saturation mechanism, it achieves a balance of
interpretability, tractability, and realism, making it a promising
candidate for both theoretical analysis and applied traffic
engineering~\cite{Wilson2011}.
\section{Conclusion}
The Hybrid OVD model captures driver behavior in a way that is
both interpretable and physically grounded. By combining the headway
response of the classical OVM with a drag-inspired acceleration law, it
retains the mechanisms that generate stop-and-go waves while ensuring
bounded accelerations and smooth convergence to equilibrium. In doing
so, the model resolves a central limitation of the classical OVM and
strengthens the bridge between behavioral traffic models and vehicle
dynamics.
Looking ahead, the hybrid formulation provides a compact and flexible
framework for both theoretical analysis and practical applications. Its
parameters relate naturally to driver heterogeneity and vehicle
capabilities, making it well suited for calibration with empirical data.
The model also lends itself to large-scale simulations and opens
opportunities for studying mixed traffic with human drivers and
automated vehicles, multi-lane interactions, and control strategies such
as adaptive cruise. The Hybrid OVD model offers a practical and interpretable framework that can support future studies in traffic flow dynamics.
\appendix
\section{Extended unstable trajectories}
\label{app:longtraj}
To understand the nonlinear growth of stop-and-go oscillations, we
extended the unstable simulation to $t=300$. The longer time window
makes the persistence of traffic jams more evident and shows the
traveling stop-and-go pattern more clearly.
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{hybrid_trajectories_long_unstable.pdf}
\caption{Extended unstable simulation of the Hybrid OVD model
with $N=100$ vehicles, ring length $L=50$ ($b=0.5$), and
$a/a^\ast=0.50$. Extending the horizon to $T=300$ highlights how
small perturbations grow into persistent stop-and-go waves.}
\label{fig:longtraj}
\end{figure}
\section{Jerk in the OVM and Hybrid OVD Model}
\label{app:jerk-analysis}
We compute the \emph{jerk}, defined as the time derivative of acceleration,
\[
j = \frac{d^2 v}{dt^2},
\]
for both the classical OVM and the Hybrid OVD model. To keep the
analysis transparent, we fix the desired velocity at a constant value
$V$, which allows us to highlight the internal structure of each model.
\paragraph{Classical OVM}
The acceleration law is
\[
\dot v = \alpha (V - v),
\]
so that
\[
j_{\text{OVM}} = -\alpha^2 (V - v).
\]
Thus jerk is linear in the deviation from $V$: negative for $v<V$,
positive for $v>V$, and with magnitude proportional to $|v-V|$.
\paragraph{Hybrid OVD model}
Here the acceleration law is
\[
\dot v = a\left(1-\frac{v^2}{V^2}\right),
\]
which yields
\[
j_{\text{Hybrid}} = -\frac{2a}{V^2} v \dot v
= -\frac{2a^2}{V^2} v\left(1-\frac{v^2}{V^2}\right).
\]
The jerk is cubic in $v$: it vanishes at $v=0$ and $v=V$, is negative
for $0<v<V$, and becomes positive for $v>V$.
\begin{figure}[H]
\centering
\includegraphics[width=0.9\textwidth]{OVM_vs_Hybrid_jerk_V20_uptoV.pdf}
\caption{Jerk $j=d^2v/dt^2$ in the OVM and Hybrid OVD model
for $V=20$ m/s, $a=5.0~\mathrm{m/s^2}$, and matched sensitivity
$\alpha=2a/V=0.5~\mathrm{s}^{-1}$. The OVM produces a linear jerk
law, while the Hybrid model yields a cubic law that vanishes at
$v=0$ and $v=V$ and remains bounded for $0 \leq v \leq V$. For
$v>V$, however, the Hybrid jerk grows cubically and can exceed that
of the OVM.}
\label{fig:ovm-hybrid-jerk}
\end{figure}
\paragraph{Note}
Near equilibrium both models behave similarly: jerk changes sign
depending on whether $v$ is above or below $V$. Globally, however, the
OVM jerk grows linearly without bound as $|v-V|$ increases, while the
Hybrid jerk is bounded within the realistic range $0 \leq v \leq V$ but
grows cubically for $v>V$. This overspeed behavior is discussed as a
modeling limitation in Section~\ref{sec:limitations}.
\bibliographystyle{IEEEtran}
\bibliography{main}
\end{document}
% Generated by IEEEtran.bst, version: 1.14 (2015/08/26)
\begin{thebibliography}{10}
\providecommand{\url}[1]{#1}
\csname url@samestyle\endcsname
\providecommand{\newblock}{\relax}
\providecommand{\bibinfo}[2]{#2}
\providecommand{\BIBentrySTDinterwordspacing}{\spaceskip=0pt\relax}
\providecommand{\BIBentryALTinterwordstretchfactor}{4}
\providecommand{\BIBentryALTinterwordspacing}{\spaceskip=\fontdimen2\font plus
\BIBentryALTinterwordstretchfactor\fontdimen3\font minus \fontdimen4\font\relax}
\providecommand{\BIBforeignlanguage}[2]{{%
\expandafter\ifx\csname l@#1\endcsname\relax
\typeout{** WARNING: IEEEtran.bst: No hyphenation pattern has been}%
\typeout{** loaded for the language `#1'. Using the pattern for}%
\typeout{** the default language instead.}%
\else
\language=\csname l@#1\endcsname
\fi
#2}}
\providecommand{\BIBdecl}{\relax}
\BIBdecl
\bibitem{Bando1995}
M.~Bando, K.~Hasebe, A.~Nakayama, A.~Shibata, and Y.~Sugiyama, ``Dynamical model of traffic congestion and numerical simulation,'' \emph{Physical Review E}, vol.~51, no.~2, pp. 1035--1042, 1995.
\bibitem{Helbing2001}
D.~Helbing, ``Traffic and related self-driven many-particle systems,'' \emph{Reviews of Modern Physics}, vol.~73, no.~4, pp. 1067--1141, 2001.
\bibitem{Nagatani2002}
T.~Nagatani, ``The physics of traffic jams,'' \emph{Reports on Progress in Physics}, vol.~65, no.~9, pp. 1331--1386, 2002.
\bibitem{Kerner2004}
B.~S. Kerner, \emph{The Physics of Traffic}.\hskip 1em plus 0.5em minus 0.4em\relax Springer, 2004.
\bibitem{Wilson2011}
R.~E. Wilson and J.~A. Ward, ``Car-following models: fifty years of linear stability analysis – a mathematical perspective,'' \emph{Transportation Planning and Technology}, vol.~34, no.~1, pp. 3--18, 2011.
\bibitem{Jiang2001}
R.~Jiang, Q.-S. Wu, and Z.-J. Zhu, ``Full velocity difference model for a car-following theory,'' \emph{Physical Review E}, vol.~64, no.~1, p. 017101, 2001.
\bibitem{Treiber2000}
M.~Treiber, A.~Hennecke, and D.~Helbing, ``Congested traffic states in empirical observations and microscopic simulations,'' \emph{Physical Review E}, vol.~62, no.~2, pp. 1805--1824, 2000.
\bibitem{Chowdhury2000}
D.~Chowdhury, L.~Santen, and A.~Schadschneider, ``Statistical physics of vehicular traffic and some related systems,'' \emph{Physics Reports}, vol. 329, no. 4-6, pp. 199--329, 2000.
\bibitem{Herman1997}
R.~Herman and S.~Ardekani, ``Dynamic traffic flow models with different car-following patterns,'' \emph{Transportation Research Part B}, vol.~31, no.~6, pp. 421--431, 1997.
\bibitem{Zhang1999}
H.~M. Zhang, ``Nonlinear stability and bifurcation in a car-following model with reaction-time delay,'' \emph{Transportation Research Part B}, vol.~33, no.~6, pp. 399--414, 1999.
\bibitem{Li2012}
X.~Li, H.~Kuang, T.~Song, Y.~Wang, and C.~Ou, ``Modified optimal velocity model with consideration of the drivers’ forecast effect,'' \emph{Physica A: Statistical Mechanics and its Applications}, vol. 391, no.~4, pp. 1503--1512, 2012.
\bibitem{Ge2004}
H.~Ge, S.~Ma, and B.~Han, ``Modified optimal velocity model considering the jam density,'' \emph{Physics Letters A}, vol. 329, no. 5-6, pp. 344--353, 2004.
\bibitem{Komatsu1995}
T.~S. Komatsu and S.~I. Sasa, ``Kink soliton solutions of the modified korteweg–de vries equation describing traffic congestion,'' \emph{Physical Review E}, vol.~52, no.~6, pp. 5574--5582, 1995.
\bibitem{Sugiyama2008}
Y.~Sugiyama, M.~Fukui, M.~Kikuchi, K.~Hasebe, A.~Nakayama, K.~Nishinari, D.~Takahashi, and S.~Yukawa, ``Traffic jams without bottlenecks—experimental evidence for the physical mechanism of the formation of a jam,'' \emph{New Journal of Physics}, vol.~10, no.~3, p. 033001, 2008.
\bibitem{Orosz2010}
G.~Orosz, R.~E. Wilson, and G.~St{\'e}p{\'a}n, ``Traffic jams: dynamics and control,'' \emph{Philosophical Transactions of the Royal Society A}, vol. 368, no. 1928, pp. 4455--4479, 2010.
\bibitem{Treiber2013}
M.~Treiber and A.~Kesting, \emph{Traffic Flow Dynamics: Data, Models and Simulation}.\hskip 1em plus 0.5em minus 0.4em\relax Berlin, Heidelberg: Springer, 2013.
\bibitem{Batchelor1967}
G.~K. Batchelor, \emph{An Introduction to Fluid Dynamics}.\hskip 1em plus 0.5em minus 0.4em\relax Cambridge University Press, 1967.
\bibitem{Wilson2008}
R.~E. Wilson, ``Mechanisms for spatio-temporal pattern formation in highway traffic models,'' \emph{Philosophical Transactions of the Royal Society A}, vol. 366, no. 1872, pp. 2017--2032, 2008.
\bibitem{Seibold2013}
B.~Seibold, M.~R. Flynn, A.~R. Kasimov, and R.~R. Rosales, ``Constructing set-valued fundamental diagrams from jamiton solutions in second order traffic models,'' \emph{Networks and Heterogeneous Media}, vol.~8, no.~3, pp. 745--772, 2013.
\bibitem{Milanes2014}
V.~Milanés and S.~E. Shladover, ``Modeling cooperative and autonomous adaptive cruise control dynamic responses using experimental data,'' \emph{Transportation Research Part C: Emerging Technologies}, vol.~48, pp. 285--300, 2014.
\bibitem{Kesting2010}
A.~Kesting, M.~Treiber, and D.~Helbing, ``Enhanced intelligent driver model to access the impact of driving strategies on traffic capacity,'' \emph{Philosophical Transactions of the Royal Society A}, vol. 368, no. 1928, pp. 4585--4605, 2010.
\bibitem{Punzo2016}
V.~Punzo, M.~Montanino, and B.~Ciuffo, ``Calibration and assessment of car-following models using ngsim trajectory data,'' \emph{Transportation Research Part C: Emerging Technologies}, vol.~73, pp. 14--31, 2016.
\bibitem{Strogatz2015}
S.~H. Strogatz, \emph{Nonlinear Dynamics and Chaos: With Applications to Physics, Biology, Chemistry, and Engineering}, 2nd~ed.\hskip 1em plus 0.5em minus 0.4em\relax Westview Press, 2015.
\bibitem{Hirsch2012}
M.~W. Hirsch, S.~Smale, and R.~L. Devaney, \emph{Differential Equations, Dynamical Systems, and an Introduction to Chaos}, 3rd~ed.\hskip 1em plus 0.5em minus 0.4em\relax Academic Press, 2012.
\bibitem{Khalil2002}
H.~K. Khalil, \emph{Nonlinear Systems}, 3rd~ed.\hskip 1em plus 0.5em minus 0.4em\relax Prentice Hall, 2002.
\bibitem{Rahman2022}
N.~Rahman, R.~Marangell, and D.~Oelz, ``Classification and stability analysis of polarising and depolarising travelling wave solutions for a model of collective cell migration,'' \emph{Applied Mathematics and Computation}, vol. 421, p. 126954, 2022.
\end{thebibliography}