A Hierarchical Spatio-Temporal Trajectory Planning Framework for Autonomous Vehicles Incorporating Road Network Topology

Motion planning for non-holonomic dynamic systems presents substantial challenges, particularly in ensuring conflict-free operation. Most existing spatio-temporal decoupling trajectory planning methods address the problem by separately optimizing the path and velocity components to prune the feasible trajectory space. However, this approach overlooks the interaction between the two dimensions, resulting in limited responsiveness to dynamic obstacles. To overcome this limitation, this paper proposes a hierarchical 3D spatio-temporal coupled trajectory planning framework which incorporates road network topology. The first layer constructs a 3D spatio-temporal corridor for specified time horizons using vector topology maps and sensor perception data to define the trajectory sampling space. In the second layer, a reference trajectory is derived within the state space using the vehicle kinematic model via sampling and graph search techniques. To improve optimization efficiency, a variable step-size trajectory smoothing strategy is introduced, leveraging road information and the vehicle's motion state to prioritize high-value targets. Furthermore, to mitigate instability arising from sensor and chassis control errors, a smoothing and attachment strategy for adjacent segments is devised to refine trajectories across consecutive frames. Simulation tests on the CARLA platform and real-world experiments demonstrate that the proposed algorithm effectively responds to dynamic obstacles, satisfies real-time requirements, and exhibits strong adaptability across diverse scenarios.