Traffic Control via Fleets of Connected and Automated Vehicles

Chiara Daini, Maria Laura Delle Monache, Paola Goatin, Antonella Ferrara

TL;DR

This paper demonstrates how a tiny fleet of Connected and Automated Vehicles (CAVs) can act as "moving bottlenecks" to smooth out traffic flow and significantly reduce fuel consumption. By testing centralized and decentralized control strategies through a multi-scale simulation, the authors show that just five optimally placed CAVs can reduce total fuel consumption by up to 6%, even without global coordination.

What Problem Does This Paper Address?

Traffic congestion isn't just a nuisance; it’s a major source of fuel waste and CO2 emissions. Much of this inefficiency comes from "stop-and-go" waves—phantom traffic jams where vehicles repeatedly accelerate and brake.

Traditional traffic management relies on "exogenous" infrastructure like ramp meters or variable speed limit signs. These are expensive to install and maintain. This paper explores an "endogenous" alternative: using CAVs already on the road as actuators. By precisely controlling the speed of a few automated vehicles, we can influence the "bulk flow" of human-driven cars behind them, effectively dampening stop-and-go waves before they escalate.

Methods

The authors employ a multi-scale modeling approach to simulate the complex interaction between individual automated vehicles and the surrounding mass of human-driven traffic:

  • Macroscopic Level: The bulk traffic is modeled using the Lighthill-Whitham-Richards (LWR) model, a first-order Partial Differential Equation (PDE) that treats traffic like a fluid.
  • Microscopic Level: Individual CAVs are tracked using Ordinary Differential Equations (ODEs).
  • Coupling: The CAVs act as "moving bottlenecks," meaning they impose a local capacity constraint on the PDE based on their current speed and position.

The study compares three control architectures:

  1. Centralized: A "god-eye" controller manages all CAVs simultaneously with full knowledge of the road.
  2. Decentralized: Each CAV optimizes its own speed based only on its local environment.
  3. Quasi-decentralized: CAVs coordinate with others within a specific sensing radius.

These are tested using both Global Optimization (assuming perfect future knowledge) and Model Predictive Control (MPC), which uses a receding horizon to make decisions in a way that is realistic for real-world deployment.

Datasets

The paper does not use a new real-world traffic dataset; instead, it relies on a high-fidelity simulation environment. However, it utilizes a specific fuel consumption model derived from empirical data on four vehicle types (Ford Explorer, Ford Focus, Honda Civic, and Honda Accord). This model maps vehicle speed to fuel consumption rates (Liters/hr), allowing the authors to calculate a "Total Fuel Consumption" (TFC) index for the entire road stretch.

Key Insights and Ideas

  • The Power of Few: One of the most striking insights is that you don't need a high "penetration rate" of CAVs to see benefits. In a 50km stretch of highway, a fleet of just 5 to 10 vehicles was enough to stabilize the flow.
  • Diminishing Returns: The gain in fuel efficiency increases with the number of CAVs but stabilizes quickly. Adding more than five vehicles provided only marginal extra benefits, suggesting that strategic placement is more important than sheer numbers.
  • Decentralized Success: Surprisingly, the decentralized and quasi-decentralized strategies performed nearly as well as the centralized one, especially under the MPC framework. This is a major win for real-world feasibility, as it suggests we don't need a massive, low-latency communication infrastructure connecting every car to a central hub.
  • Strategic Positioning: The "sensitivity analysis" showed that CAVs have the most impact when they start just downstream of high-density traffic regions, allowing them to "catch" and smooth out waves early.

Key Results

  • Fuel Savings: Under global optimization, a fleet of 5 CAVs achieved a 6.14% reduction in total fuel consumption compared to an uncontrolled scenario.
  • MPC Performance: In the more realistic MPC setup (which accounts for computational lag and limited look-ahead), the fuel reduction was approximately 3.7% to 3.8%.
  • Stability: The control strategies successfully dissipated stop-and-go waves, as evidenced by the spatio-temporal density plots which showed much smoother transitions in the controlled cases compared to the "red zones" of congestion in the uncontrolled baseline.

Limitations and Future Work

The authors acknowledge several simplifications that provide clear paths for future research:

  • Constant Speed Assumption: To keep the optimization computationally tractable, the CAVs were assigned constant optimal speeds within each MPC horizon. Future work could explore time-varying speed profiles for even finer control.
  • First-Order Model: The LWR model is "first-order," meaning it doesn't explicitly account for vehicle acceleration. Since acceleration is a major fuel consumer, moving to a "second-order" model might reveal even greater potential for savings.
  • No Lane Changing: The current model assumes CAVs stay in their lanes (though they can be overtaken by others). Incorporating active lane-changing as a control variable could allow CAVs to "block" or "filter" traffic more effectively.
  • Human Factor: The model assumes human drivers follow a standard speed-density law. Real-world human behavior (like aggressive cutting-in) might impact the effectiveness of the moving bottlenecks.

Summary updated

Abstract

In this paper, we propose three control strategies, based on different levels of cooperation (centralized, decentralized and quasi-decentralized), to improve density dependent traffic performance indexes, such as fuel consumption, by acting on a small number of Connected and Automated Vehicles (CAVs) operating as moving bottlenecks on the surrounding flow. We rely on a multi-scale approach to model mixed traffic of CAVs in the bulk flow. In particular, CAVs are individually tracked and they are allowed to overtake (if on distinct lanes) or queuing (if on the same lane). Controlling CAVs desired speeds allows to act on the system to minimize the selected cost function. For the proposed control strategies, we apply both global optimization and a Model Predictive Control approach. In particular, we perform numerical tests to investigate how the CAVs number and positions impact the result, showing that few, optimally chosen vehicles are sufficient to significantly improve the selected performance indexes, even using a decentralized control policy. Simulation results support the attractive perspective of exploiting a very small number of vehicles as endogenous control actuators to regulate traffic flow on road networks, providing a flexible alternative to traditional control methods.
BibTeX
@article{daini2024traffic,
  author = {Chiara Daini and Maria Laura Delle Monache and Paola Goatin and Antonella Ferrara},
  title = {{Traffic Control via Fleets of Connected and Automated Vehicles}},
  journal = {IEEE Transactions on Intelligent Transportation Systems},
  year = {2024},
  doi = {10.1109/tits.2024.3506703},
}
Figure