Driving AI: Centralized Decision and Control for Multiple Vehicles at Crossroad by RL

This project aims to develop a centralized coordination scheme of automated vehicles at an intersection without traffic signals using RL to address low computation efficiency suffered by current centralized coordination methods. My works: 1) Proposed model accelerated proximal policy optimization algorithm, which incorporates a prior model into PPO to facilitate sample efficiency. 2) Designed state with minimal length, and simplified reward function under consideration of safety, efficiency and task completion. The computing efficiency and traffic efficiency are respectively 400 times and 4.5 times higher that those of the baseline method based on MPC.