CIRCLES: Congestion Impacts Reduction via CAV-in-the-loop Lagrangian Energy Smoothing

Alexandre M. Bayen, Jonathan Lee, Benedetto Piccoli, Benjamin Seibold, Jonathan Sprinkle, Daniel B. Work2024 · DOI

Abstract

The energy efficiency of today’s vehicular mobility relies on the un-integrated combination of static infrastructure control (e.g., traffic lights) and onboard vehicle automation (e.g., adaptive cruise control). These systems are not co-designed, often leading to traffic instabilities and "phantom" jams. The CIRCLES project introduces Mobile Traffic Control (MTC), where automated vehicles actuate traffic flow to optimize the energy footprint. Through a live-traffic demonstration involving 100 Connected and Automated Vehicles (CAVs) on a Tennessee highway, the project demonstrated that a local penetration rate of just 2% can significantly reduce stop-and-go oscillations and improve throughput.

Introduction: Beyond Static Infrastructure

Traditional traffic management is fundamentally "Eulerian"—fixed in space. Tools like traffic lights, ramp meters, and variable speed limit signs attempt to regulate flow from the sidelines. Simultaneously, modern vehicles utilize Adaptive Cruise Control (ACC) to manage individual longitudinal dynamics. However, these two paradigms often work at cross-purposes. Research has shown that standard ACC can actually amplify traffic waves, creating "phantom jams" that waste fuel and increase travel time.

The CIRCLES project proposes a shift toward Mobile Traffic Control (MTC). In this "Lagrangian" approach, the controllers move with the traffic flow. By using a small fleet of CAVs as actuators, the system can smooth out instabilities from within the traffic stream itself, effectively "calming" the surrounding human-driven vehicles.

The MegaController Architecture

To achieve coordinated traffic smoothing, the team developed a hierarchical control system known as the MegaController. This architecture bridges the gap between high-level traffic management and low-level vehicle actuation through two distinct layers:

  1. Centralized Speed Planner: This layer serves as the system's "brain." It ingests real-time data from third-party traffic providers and the CAVs' own telemetry. Using Traffic State Estimation (TSE), the Speed Planner calculates optimal target speeds designed to dissipate downstream congestion before it fully forms.
  2. Decentralized Local Control Layer: This layer resides on the individual vehicles. It overrides the stock ACC system to implement the Speed Planner’s targets. Because the fleet was heterogeneous (comprising vehicles from three different manufacturers), this layer had to be hardware-agnostic, translating high-level commands into specific longitudinal torque or acceleration requests.

The development of these controllers relied on Deep Reinforcement Learning (RL). To make RL training feasible, the team created a model reduction pipeline that simplified complex vehicle dynamics into "physics-like" models. These models remained energy-accurate while being computationally efficient enough for real-time optimization.

The MegaVandertest: Validating MTC at Scale

The project culminated in the MegaVandertest, a massive five-day field experiment conducted in November 2022. The test took place on a six-mile stretch of the I-24 highway near Nashville, Tennessee.

The scale of the deployment was unprecedented:

  • Fleet: 100 specially-equipped CAVs from three manufacturers.
  • Operations: Over 1,022 hours of driving and 22,752 miles of real-world telemetry.
  • Monitoring: The experiment utilized the I-24 MOTION system, a sophisticated sensor array capable of tracking every vehicle on the roadway, providing a ground-truth dataset for validation.

This test aimed to prove the "2% Solution": the hypothesis that a very low penetration rate of automated vehicles is sufficient to influence the behavior of the entire traffic stream.

Key Insights and Results

The results from the MegaVandertest confirmed the viability of MTC for energy and flow optimization. By treating traffic as a fluid and reducing "turbulence" (speed variance), the system improved the overall efficiency of the roadway.

Key quantitative findings include:

  • Stability Gains: The standard deviation of vehicle speeds was reduced by 34.36%. This indicates a massive reduction in stop-and-go waves, leading to a more predictable and less stressful driving environment.
  • Throughput Improvement: Bottleneck throughput increased by 5.01%, allowing more vehicles to pass through the same stretch of highway in less time.
  • Scalability: The successful management of 100 vehicles in high-density, live traffic demonstrated that RL-based controllers can transition from simulation to complex real-world environments.

Conclusion and Future Directions

The CIRCLES project represents a significant departure from traditional vehicular control. By demonstrating that a 2% penetration of CAVs can actuate the flow of thousands of human-driven cars, the project opens new doors for energy-efficient mobility.

While the traffic smoothing results are definitive, the team is continuing to refine the quantification of total fuel savings across the entire traffic stream. Future research will focus on scaling the fleet further, adapting MTC for complex urban environments with intersections, and integrating these socially aware control algorithms into commercial autonomous vehicle stacks.

Summary updated

Abstract

The energy efficiency of today’s vehicular mobility relies on the un-integrated combination of i) control via static assets (traffic lights, metering, variable speed limits, etc.); and ii) onboard vehicle automation (adaptive cruise control (ACC), ecodriving, etc.). These two families of control were not co-designed and are not engineered to work in coordination. Recent studies have shown i) limitations of controls, and even ii) negative impacts of ACC [1]. This project focused on the technology development, implementation and prototyping, and validation of Mobile Traffic Control (MTC). MTC can be viewed as an extension of classical traffic control (in which static infrastructure actuates traffic flow). In the MTC paradigm, automated vehicles actuate the entire flow via their behavior, offering enhanced possibilities to optimize the energy footprint of traffic, if designed correctly. We set out to demonstrate for the first time that considerably reduced fuel consumption of all vehicles in traffic can be achieved via distributed control of a small proportion of CAVs. Compared to baseline vehicular technologies, our work offers a significant design departure: control algorithms for the CAVs consider the impact one vehicle can have on overall traffic, improving resulting overall fuel consumption. We focus on using a few vehicles (as traffic controllers via CAV technology) to improve the energy efficiency of traffic flow to further optimize energy efficiency. A live-traffic demonstration in November 2022 featured the deployment of 100 specially-equipped CAVs, constituting an approximate local penetration rate upwards of 2%.
BibTeX
@misc{bayen2024circles,
  author = {Alexandre M. Bayen and Jonathan Lee and Benedetto Piccoli and Benjamin Seibold and Jonathan Sprinkle and Daniel B. Work},
  title = {{CIRCLES: Congestion Impacts Reduction via CAV-in-the-loop Lagrangian Energy Smoothing}},
  year = {2024},
  doi = {10.2172/2371573},
}
Figure