Simulation Study on Cooperative Control of Traffic Signals and Connected and Automated Vehicles Found Up to a 30 Percent Reduction in Travel Time Compared to Baseline.
Simulation Testing Demonstrated the Effectiveness of an AI-Based Signal Control System in Three Scenarios, Including a Realistic Urban Scenario of Dublin, Ireland.
Dublin, Ireland
CoTV: Cooperative Control for Traffic Light Signals and Connected Autonomous Vehicles Using Deep Reinforcement Learning
Summary Information
Most research in urban traffic control adjusts either traffic signalsto improve the intersection throughput or vehicle speed to stabilize the traffic, rather than joint control. In this study, researchers tested an AI-based control algorithm, CoTV, based on multi-agent Deep Reinforcement Learning (DRL) to balance the advantages of both in two hypothetical scenarios and one scenario using real world data. To scale to complex urban scenarios, the system selected the closest connected and automated vehicle (CAV) to the intersection on each incoming road as the agent. The way these agents sense their environment and receive feedback was simplified to make communication more efficient and the system easier to deploy.
METHODOLOGY
This simulation study assumed that all vehicles are connected. The proposed system components included traffic light controller agents, CAV agents, and a DRL training process, all designed to be deployed in existing adaptive traffic light systems.
- Three testing scenarios in the microsimulation platform included: Scenario 1) a 1x1 grid map with a single intersection, Scenario 2) a 1x6 grid map, and Scenario 3) six signalized intersections in the city of Dublin.
- The researchers compared CoTV to a baseline static timing plan, two other DRL-based models, and a non-DRL method for jointly controlling traffic signals and CAVs.
- Evaluation metrics included travel time, delay, fuel consumption, and time-to-collision (time until a potential crash if speed and direction stay the same).
FINDINGS
- Compared to the baseline scenario, CoTV under a 100 percent CAV penetration rate achieved:
- The shortest travel time, with reductions of 18.85 percent (Scenario 1), 27.15 percent (Scenario 2), and 29.61 percent (Scenario 3).
- Reductions in delay from 60.02 to 73.74 percent.
- Reductions in fuel consumption from 9.15 to 27.41 percent.
- Reductions in time-to-collision from 83.84 to 96.04 percent.
- The travel time of CoTV tended to decrease as the penetration rate increased for all three scenarios.
