Abstract:

A trajectory planning is proposed for collision avoidance in a corner of intelligent vehicle considering trajectory smoothness and vehicle stability. The trajectory planning is decomposed into path planning and speed planning. The improved rapidly exploring random tree （RRT） is used to construct a collision-free continuous-curvature clothoids with minimum curvature change. Based on the measurement function of deep neural network， the improved RRT selects and connects the tree nodes with the smallest cost function， and searches the nearby nodes to find whether there are nodes with smaller cost function near the selected nodes. In speed planning， trapezoidal velocity profiles are used to output continuous target acceleration curve according to the road speed limit rules. Then， based on the curvature of clothoids and vehicle state， the target acceleration is dynamically adjusted by Preview G-Vectoring control （PGVC）. Finally， the final expected acceleration is obtained through acceleration control logic. The simulation results show that the proposed trajectory planning method can not only achieve collision avoidance in a corner and guarantee a good tracking performance， but also improve the stability in a corner at high speed. Besides， the paper also verifies the fast convergence， path smoothness and real-time performance based on parallel computing of RRT.

Key words: intelligent vehicle, trajectory planning, rapidly exploring random tree, speed planning, cornering collision avoidance