Monte Carlo Tree Search Based Trajectory Generation for Automated Vehicles in Interactive Traffic Environments

This paper focuses on the development of a trajectory planning method for connected and automated vehicles (CAVs) that takes into account the interactive nature of the vehicles. The proposed approach is based on Monte Carlo Tree Search (MCTS) that traverse through possible actions from each state of the system to identify the trajectory with highest reward. Here, the trajectory is planned and the actions of surrounding vehicles are predicted jointly. Planning the trajectory and predicting the surrounding vehicles jointly in an interactive environment can result in a large action-space, which is not computationally tractable. Hence, we propose an adaptive action-space, which includes pruning the action-space so that the actions resulting in unsafe trajectories are eliminated. The simulation studies show that the proposed approach is capable of identifying less conservative yet safe trajectories for CAVs in a multi-vehicle environment.