⚡ electrical engineering

A Continuification Approach to CAV Control in Mixed Traffic via Variable Speed Limits

This paper proposes a computationally efficient control strategy for multiple connected and automated vehicles (CAVs) in mixed traffic that designs an optimal variable speed limit policy using a PDE-based LQR approach and subsequently derives individual CAV speeds through a continuification method, demonstrating that the solution converges to the global optimum as the number of CAVs increases.

Original authors: Brian Block, Cecilia Pasquale, Silvia Siri, Simona Sacone, Stephanie Stockar

Published 2026-06-09

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Brian Block, Cecilia Pasquale, Silvia Siri, Simona Sacone, Stephanie Stockar

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a highway as a long, flowing river. Usually, cars move like water, speeding up when the river is wide and empty, and slowing down when it gets crowded. Sometimes, a "traffic jam" is like a rock in the river, creating a ripple of slow-moving water that travels backward, causing chaos for everyone behind it.

This paper proposes a new way to smooth out those ripples using Connected and Automated Vehicles (CAVs). Think of these CAVs as "smart rafts" floating in the river. When they slow down or speed up, they act like a moving bottleneck, forcing the cars around them to adjust their speed, which can help dissolve traffic jams.

Here is the simple breakdown of what the authors did and found:

The Problem: Too Many Rafts, Too Much Math

Traditionally, to control these "smart rafts," engineers tried to calculate the perfect speed for each individual raft one by one.

The Old Way (ODE-based): Imagine trying to calculate the perfect speed for 1, 2, or 5 rafts by solving a complex math puzzle for every single one separately. If you have just one raft, it's manageable. But if you add more rafts, the math puzzle gets huge and takes a long time to solve. It's like trying to direct a parade by shouting instructions to every single marcher individually; as the parade grows, you get overwhelmed.

The New Idea: The "Big Picture" Map

The authors came up with a clever shortcut called "continuification."

The New Way (PDE-based): Instead of looking at every single raft individually, they first looked at the entire river as a whole. They created a "Big Picture Map" (a mathematical model called a PDE) that tells them the perfect speed limit for the entire length of the highway to smooth out the water.
The Translation: Once they have this perfect "Big Picture Map," they simply look at where their specific "smart rafts" (CAVs) are located and tell them, "Hey, go at the speed the map says is right for this spot."

It's like a conductor directing an orchestra. Instead of telling every violinist exactly how to play their specific note (which takes forever), the conductor sets the tempo for the whole song, and the musicians just follow the beat appropriate for their section.

What They Tested

They ran computer simulations of a circular highway with different numbers of "smart rafts" (1, 2, and 5) to see which method worked better.

The Goal: Get the traffic density (how crowded the road is) to a smooth, steady level.
The Comparison: They compared their new "Big Picture Map" method against the old "shout at every raft" method (which used a complex optimization tool called MPC).

The Results

Speed and Efficiency: The new method was incredibly faster. In the case of 5 rafts, the new method was 600 times faster than the old method. The old method actually became too slow to be useful in real-time when there were multiple rafts, while the new method stayed fast and efficient no matter how many rafts were added.
Performance: Both methods could eventually clear the traffic jams, but the new method cleared them much quicker. As they added more rafts, the new method's performance got closer and closer to the "perfect theoretical solution" (the absolute best possible outcome).
Realism: The authors noted that sometimes the "Big Picture Map" tells a raft to go faster than the surrounding traffic allows. In those cases, the raft just follows the traffic speed (it can't fly!). This happens, but the system handles it gracefully.

The Bottom Line

The paper claims that by designing the control strategy for the whole highway first and then just "plugging in" the specific vehicles, we can control traffic much more efficiently. It allows us to use many automated vehicles to fix traffic jams without getting bogged down in slow, heavy math calculations.

In short: Instead of micromanaging every car, they figured out how to manage the whole road and let the smart cars just follow the flow, resulting in a system that is both faster to compute and better at clearing traffic.

Technical Summary: A Continuification Approach to CAV Control in Mixed Traffic via Variable Speed Limits

Problem Statement
Reducing traffic congestion and regulating system density toward a desired profile is a primary objective of modern traffic control strategies. While Variable Speed Limit (VSL) control and ramp metering are established methods, the use of Connected and Automated Vehicles (CAVs) as control actuators has gained prominence. Current approaches for controlling CAVs in mixed traffic typically model their influence as "moving bottlenecks" using a coupled Partial Differential Equation-Ordinary Differential Equation (PDE-ODE) framework. In this framework, the PDE describes the macroscopic traffic flow (density and flow), while the ODE models the specific dynamics of the moving CAV.

Existing control strategies for these coupled models are generally designed directly on the ODE constraint (e.g., using Proportional-Integral controllers or Model Predictive Control). A significant limitation of this approach is computational scalability: as the number of CAVs increases, the dimensionality of the control problem grows, leading to increased computation times that may hinder real-time implementation. Conversely, designing control directly on the macroscopic PDE is computationally efficient but often assumes that every agent in the system is controllable, which is not valid for sparse CAV penetration. This paper addresses the gap between these two extremes by seeking a method to control a sparse number of CAVs using a PDE-based design that remains computationally efficient.

Methodology
The proposed methodology utilizes a "continuification" approach to bridge the gap between macroscopic PDE control and microscopic CAV actuation. The process involves three main stages:

Macroscopic PDE Control Design: The authors start with the Lighthill-Whitham-Richards (LWR) model, a first-order scalar hyperbolic PDE describing traffic density. They utilize a previously derived infinite-dimensional Linear Quadratic Regulator (LQR) designed for VSL control. This controller determines the optimal speed limit profile across the entire length of the freeway to drive traffic density to a desired setpoint. The control input is derived analytically by solving an algebraic Riccati equation on the Hilbert space of the system.
Continuification and Discretization: Instead of designing a controller for each individual CAV, the optimal infinite-dimensional VSL control signal is discretized to the specific locations of the available CAVs. The control action $u_i(t)$ for a specific CAV $i$ at position $z_i$ is assigned the value of the optimal VSL profile at that location.
Actuation via Moving Bottleneck Dynamics: The assigned speed serves as the desired velocity for the CAV. However, the actual velocity is constrained by the surrounding traffic flow. If the commanded speed exceeds the speed of the surrounding traffic (or if the density is too high), the CAV acts as a moving bottleneck, adapting its speed to the local traffic conditions as defined by the coupled PDE-ODE model.

This approach is contrasted with a baseline ODE-based Model Predictive Control (MPC) strategy, which solves a constrained optimization problem at each time step to minimize the Root Mean Square Error (RMSE) of the traffic density, explicitly accounting for the dynamics of each individual CAV.

Key Contributions

Novel Control Architecture: The paper introduces a PDE-based moving bottleneck controller that avoids the computational burden of ODE-based multi-agent control. By designing the controller on the infinite-dimensional PDE first and then mapping it to agents, the method decouples the control complexity from the number of CAVs.
Computational Efficiency: The proposed method demonstrates that the control input for any number of CAVs can be computed via a single analytical solution to the PDE control problem, followed by simple point-wise evaluation. This contrasts with ODE-based MPC, where the optimization problem size scales with the number of vehicles.
Convergence to Global Optimum: The study investigates the behavior of the system as the number of CAVs increases, showing that the PDE-based approach converges toward the global optimal solution determined by the infinite-dimensional LQR.

Simulation Results
The efficacy of the proposed method was evaluated using a numerical simulation of a circular road with normalized parameters. The study compared the PDE-based continuification controller against an ODE-based MPC baseline across scenarios with 1, 2, and 5 CAVs.

Performance: Both controllers successfully drove the traffic density toward the desired profile. However, the PDE-based controller consistently attenuated high-density waves faster than the ODE-based MPC. In the 5-CAV scenario, the PDE-based approach regulated density significantly quicker, approaching the performance of the ideal infinite-dimensional global optimal VSL solution.
Computational Time: The PDE-based approach was substantially more efficient. In the 5-CAV case, it was approximately 600 times faster than the ODE-based MPC. While the ODE-based MPC required computation times exceeding real-time limits (Real Time Factor > 1) for 2 and 5 CAVs, the PDE-based approach remained well within real-time capabilities (Real Time Factor < 0.005) for all cases.
Initial Conditions: The results highlighted that the initial placement of CAVs impacts the transient error profile, particularly in single-CAV scenarios, though all configurations eventually converged to the desired density.

Significance and Claims
The paper claims that the proposed continuification approach offers a superior balance between control performance and computational feasibility for mixed traffic scenarios. By shifting the control design to the macroscopic PDE level, the method achieves near-global optimal performance while maintaining low computational overhead regardless of the number of controlled CAVs. The authors assert that this makes the approach particularly suitable for large-scale traffic networks and scenarios with increasing CAV penetration rates, where ODE-based methods become computationally prohibitive. The work concludes that the PDE-based controller is not only more efficient in reaching the desired density profile but also scalable, approaching the infinite-dimensional global optimal solution as the density of controlled agents increases. Future work is identified as applying these control actions to more realistic microscopic simulators and analyzing the impact of CAV placement on controllability in non-circular road networks.

The Problem: Too Many Rafts, Too Much Math

The New Idea: The "Big Picture" Map

What They Tested

The Results

The Bottom Line

More like this