关键词: 自动驾驶车辆, 轨迹融合预测, 卡尔曼滤波器, 动态规划, 可行边界, 车辆动力学, 模型预测控制

Abstract: To address the problem that trajectory predictions for traffic participants produced by traditional motion planning algorithms are unsuitable for complex driving scenarios and are not effectively integrated with subsequent motion planning, resulting in incomplete utilization of obstacle position information, a longitudinal and lateral joint motion planning algorithm for autonomous driving is proposed based on multi-model fusion prediction of obstacle trajectories. First, the Constant Acceleration (CA) model and the Constant Turn Rate and Velocity (CTRV) model are selected respectively as the long-term and short-term prediction models for traffic participants, and the prediction results are fused using a method based on the Kalman filter. Second, the spatiotemporal occupancy within the prediction horizon is gridded, and, using the obstacle trajectories obtained from the fusion prediction, a dynamic programming algorithm is executed to obtain new feasible boundaries. Third, a Linear Time-Varying (LTV) vehicle dynamics model is established and the ego vehicle′s global trajectory is parameterized; a classical Model Predictive Control (MPC) problem is then formulated and solved via quadratic programming to achieve joint longitudinal and lateral motion planning that yields the desired collision-free motion. Finally, joint simulations are performed on a verification platform built with CarSim software and Simulink software in a three-lane driving scenario. The results indicate that the proposed longitudinal and lateral joint motion planning algorithm based on multi-model fusion prediction of obstacle trajectories can effectively integrate obstacle vehicle trajectory prediction with the ego vehicle′s joint motion generation task. The multi-model fusion prediction algorithm demonstrates faster response and smaller prediction errors in continuous lane-change scenarios, providing a reference for research on motion planning of autonomous vehicles in dynamic obstacle environments.

Key words: autonomous vehicles, trajectory fusion prediction, Kalman filtering, Dynamic Programming (DP), feasible boundary, vehicle dynamics, Model Predictive Control (MPC)