Multi-Vehicle Cooperative Inflow Trajectory Planning Based on Reinforcement Learning and Collaborative Coordinate Framework

The expressway ramp merging zone represents a classic traffic bottleneck, posing significant challenges to multi-vehicle collaborative trajectory planning. To enhance traffic efficiency while ensuring safety, this study proposes a hierarchical planning method integrating a collaborative relative coordinate framework with reinforcement learning. First, a dynamic relative coordinate system is established to uniformly describe multi-vehicle interactions through static and dynamic risk fields, decoupling complex global planning into relative spatial dimensions. Subsequently, a trajectory planner based on a depth-deterministic policy gradient algorithm is designed within this framework, featuring a composite reward function that optimizes efficiency, safety, and comfort. CARLA simulation results demonstrate that this method effectively guides vehicles to achieve smooth and safe collaborative merging, with trajectories remaining smooth and pre-collision times consistently within safe thresholds. The approach significantly outperforms traditional methods, providing innovative solutions for collaborative decision-making in mixed traffic flow scenarios.