Trajectory Planning for Autonomous Driving Featuring Time-Varying Road Curvature and Adhesion Constraints

Yifan Gao, Wei Li, Yu Hu

Abstract: Among the various driving situations, there are challenging road conditions where both the texture and curvature are variables over time (e.g., mountainous area). However, it is found that the characteristics of road texture and curvature have been respectively considered in some of the existing studies to determine the vehicle speed for trajectory planning, but the complementary effect of these two factors is still yet to be incorporated. This could lead to unsafe vehicle behaviour. This limitation has led us to develop a trajectory planning method that gives a systematic consideration of road conditions and leverages the complementary effect of road curvature and adhesion on the vehicle speed. It prioritises the trajectory safety through a preview of road constraints (i.e., waypoints, curvature and adhesion) in a look-ahead distance and the real-time computation of the vehicle speed that satisfies the constraints. In the experiment, our method was compared with the state-of-the-art techniques in a simulated mountainous driving environment, namely Model Predictive Control (MPC), Deep Reinforcement Learning (DRL) and Hybrid A*. The environment was built with abundant variation in road curvature and adhesion. The results showed that our approach was able to generate safe and comfort trajectories in both sharp turn and ice-covered driving scenarios, in which the vehicle successfully passed through the whole length of the global path without producing large deviations and exceeding lane boundaries. Whereas, the MPC, DRL and Hybrid A* approaches resulted in the vehicle exceeding lanes at some point with completeness levels of 77.72%, 75.31% and 79.53%, respectively.

External IDs:dblp:journals/tits/GaoLH24