An End-to-end Obstacle Avoidance Algorithm for Autonomous Vehicles Based on Global Path Constraint

In recent years, local obstacle avoidance algorithm for autonomous vehicles has matured significantly. Existing methods can be divided into two stages: path planning and motion planning, which have the disadvantage of high computational complexity and falling into local optimum easily. In this paper, we propose an end-to-end obstacle avoidance method based on obstacle location information and global path. Firstly, a mathematical model is constructed to transform the obstacle information into local error. Secondly, the states of the vehicle at several subsequent moments are estimated. Finally, the PD controller is utilized to control the vehicle, which considers the local error and the global attitude error. The result of our method is compared with standard DWA algorithm and the action of real human drivers regarding the path smoothness, lateral acceleration and the pose after the obstacle avoidance. The experimental result shows the advantage of our method. There is no need to select the goal point of each period with the constraint of the expected pose. We can achieve the goal with a smooth trajectory and high efficiency.