Smart Traffic Planning for Autonomous Cars at Busy Intersections

What is it about?

The study addresses the challenges of mandatory lane changes (MLCs) at intersections by formulating a joint longitudinal-lateral trajectory planning problem in mixed traffic as a multiagent reinforcement learning (MARL) task. It introduces SS-MA-PPO, a Simulation-Supervised Multi-Agent Proximal Policy Optimization framework, to guide connected and autonomous vehicles (CAVs) in both acceleration and lane-change decisions. A Simulation-Guided Supervisory Module (SGSM) is employed for offline trajectory rollouts of human-driver models to assess feasibility and safety while arbitrating between rule-based and learned policies. The approach incorporates the information of surrounding vehicles to promote vehicle cooperation, along with a transfer learning mechanism to accelerate training. Experiments using a real-world dataset from Langfang, China, demonstrate that SS-MA-PPO outperforms conventional and MARL baselines across various metrics. Ablation experiments confirm the effectiveness of the SGSM module, vehicle cooperation, and transfer learning, resulting in enhanced performance and faster training convergence.

Featured Image

Photo by Siyuan on Unsplash

Photo by Siyuan on Unsplash

Why is it important?

This study is important as it addresses the significant challenges posed by mandatory lane changes (MLCs) in mixed traffic environments, which can lead to increased vehicle delay, fuel consumption, and traffic inefficiencies. By developing a novel multiagent reinforcement learning framework, the research offers a solution that guides connected and autonomous vehicles (CAVs) in making more efficient acceleration and lane-change decisions. This approach has the potential to enhance traffic flow, reduce emissions, and improve safety at intersections, which are critical bottlenecks in urban traffic systems. The study's findings contribute to the advancement of intelligent transportation systems, paving the way for more effective integration of CAVs in real-world traffic scenarios. Key Takeaways: 1. Improved Traffic Efficiency: The study demonstrates that the proposed SS-MA-PPO framework significantly outperforms both conventional methods and existing multiagent reinforcement learning baselines, resulting in reduced vehicle delay and improved traffic flow at intersections. 2. Effective Vehicle Cooperation: The incorporation of surrounding vehicle information into the observation space allows for enhanced cooperation among vehicles, leading to more coordinated and efficient lane changes in mixed traffic environments. 3. Accelerated Training and Performance: The implementation of a Simulation-Guided Supervisory Module (SGSM) and a transfer learning mechanism substantially accelerates training convergence and enhances the overall effectiveness of trajectory planning, as verified by ablation experiments.