Trajectory planning and hierarchical robust control for autonomous driving on unstructured roads

To address the problems of inaccurate trajectory planning and reduced tracking control stability of autonomous vehicles caused by the typical characteristics of urban unstructured roads, such as severe mixed traffic among traffic participants, high randomness of traffic behaviors, poor functional continuity of road boundaries caused by the above factors, and consequent difficulty in traffic flow data collection, a collaborative optimization method based on a dynamic feasible region and hierarchical robust model predictive control was proposed to improve the safety and stability of autonomous driving. First, for trajectory planning, the road scenario was divided into multiple static feasible regions. A dynamic feasible region was constructed by integrating driving intention analysis, travel trend prediction, and the proportion of space occupied in each static feasible region. The trajectory was parameterized and modeled by piecewise Bézier curves. For tracking control, a hierarchical robust model predictive control (HRMPC) architecture was adopted. The upper layer optimized and generated a nominal system based on the vehicle dynamics model to determine the nominal state. The lower layer obtained sample trajectories by adding finite random factors. A tube constraint based on the 95% confidence interval was then constructed, and the actual system was used to optimize the actual control input. A two-layer collaborative optimization model was thus formed. The effectiveness of the method was demonstrated through proofs of robust stability and asymptotic stability. The results show that, in urban unstructured experimental scenarios, the maximum acceleration is 1.812 m s^-2 and the minimum acceleration is -2.103 m s^-2. The velocity and acceleration curves remain continuous without abrupt changes. The influence of finite random disturbances can be effectively addressed, and safe and smooth trajectories can be planned. Compared with existing robust model predictive control (RMPC) and tube-based robust model predictive control (Tube-RMPC) methods, the proposed method reduces the maximum tracking error by more than 7.4% and the sideslip angle of the center of mass by 26.3%. The results confirm that the proposed dynamic feasible region combined with HRMPC has good stability and can effectively improve the tracking control performance of autonomous vehicles in urban unstructured scenarios.