Application of Reinforcement Learning in Open Space Planner for Apollo Auto

Local planner makes a trajectory physically executable for an agent. Open Space Planner of Apollo framework based on nonlinear optimization methods smooths the trajectory received from a global planner. Such dependency on a global planner forces an agent to relaunch both planners when local changes occur (e.g., when an environment has dynamic obstacles), what can waste too much time. In this article, we consider a different approach which is based on reinforcement learning. This method allows agent generate a trajectory using information about environment (the current and goal state, lidar sensors, etc.). Experiments conducted on the simplified environment show that such algorithm can be implemented as the local planner in Apollo infrastructure.