A cooperative control method for connected and automated vehicles at unsignalized intersection based on Monte-Carlo tree search

Abstract

Enhancing the efficiency of traditional traffic light-based intersection control in intricate and dynamic traffic scenarios proves challenging due to vehicular queuing and frequent starts and stops. Addressing this challenge, a cooperative control method employing Monte-Carlo tree search (MCTS) is proposed for managing connected and automated vehicles (CAVs) at unsignalized intersection. A hierarchical framework for cooperative control of CAVs at unsignalized intersection is proposed, featuring centralized control and a space-time resource reservation mechanism. Acknowledging the presence of an unreliable V2X communication environment, the framework incorporates a time-out retransmission mechanism and a communication model based on a finite state machine. Subsequently, safety constraints are formulated to develop a trajectory planning method for CAVs using distributed MCTS. Accounting for both unbalanced traffic flow and an unreliable communication environment, four distinct test schemes are devised. Simulation tests are conducted by comparing the proposed method against the fixed time control (FTC) and actuated light control (ALC) strategies. Subsequently, a comparison is made with existing reinforcement learning methods. The results demonstrate that the proposed method consistently outperforms the other methods in all test cases. Particularly in scenarios of unbalanced traffic flow, the proposed method demonstrates a maximum average vehicle velocity 11.3 times and 5.1 times higher than that of the FTC and ALC, with an optimal vehicle density merely reaching 27.3% and 34.0% in comparison. Moreover, the average queue length consistently remains at zero. The proposed method reliably safeguards vehicle safety and efficiency across diverse unreliable communication environments, and is also capable of maintaining low vehicle delays across different traffic generation rates.