Hierarchical cooperative planning incorporating MARL guidance and interactive spatio-temporal corridor constraints

The problem of multi-vehicle cooperative planning has long faced challenges in balancing computational efficiency and trajectory optimization quality. By integrating multi-agent reinforcement learning (MARL)-based guidance with a hierarchical decoupling framework incorporating spatio-temporal corridor constraints, the complex coordination problem can be simplified into a set of single-vehicle optimization tasks, thereby lowering computational complexity while ensuring decision correctness and trajectory quality. To this end, this paper proposes a hierarchical planning approach that integrates MARL with spatio-temporal corridor-constrained optimization. The cooperative plan task is decomposed into multiple single-vehicle subproblems to reduce overall computational burden. At Decision-Guidance Layer, a trained MARL policy generates initial reference trajectories to guide the optimization process. At Trajectory Optimization Layer, spatio-temporal corridors are constructed based on the reference trajectories to decouple the planning for individual vehicles. Nonlinear optimization is then employed to generate high-quality trajectories, accounting for vehicle dynamics and spatio-temporal constraints. This approach enhances multi-vehicle interaction while significantly reducing computational costs and improving trajectory feasibility and smoothness. The proposed method is validated through simulations in general structured urban scenarios, as well as real-world vehicle experiments, demonstrating its adaptability and practical effectiveness.