A Dynamic Collaborative Planning Method for Multi-vehicles in the Autonomous Driving Platform of the DeepRacer

The main research content of this paper is vehicle longitudinal and lateral control and complex scheduling problem of vehicle platoon. For the navigation module of single vehicle control, Gazebo is used to simulate the real environment, and 2D lidar, binocular camera and tag codes are used to carry out vehicle localization and navigation. For the multi-vehicle scheduling part, this paper mainly pays attention to switching of platoon states. One is the repeated switch of the convoy from formation to disbandment, and two control strategies are adopted, namely, disbandment state control strategy and platoon state control strategy. In order to achieve seamless switching between two states, a new upper layer algorithm is proposed. Second, when the fleet is on and off the overpass, the control methods can be switched to ensure the correct navigation of the vehicles. The third is the situation that the follower of the platoon overtakes the car and becomes the leader or follower in the middle of the platoon fall back to become a follower at the back of the line. The simulation is in Gazebo using DeepRacer in the Amazon's platform.