Original title: How to ensure a safe control strategy? Towards a SRL for urban transit autonomous operation
Authors: Zicong Zhao
This article explores the use of deep reinforcement learning in urban rail transit autonomous operation. While deep reinforcement learning has shown promise in decision-making, its inability to guarantee safety during learning and execution has hindered its practical application in this context. To address this challenge, the article proposes a framework called SSA-DRL. This framework combines linear temporal logic, reinforcement learning, and Monte Carlo tree search. It consists of four modules: a post-posed shielding, a searching tree module, a DRL framework, and an additional actor. The output of the framework ensures speed constraint, schedule constraint, and optimization of the operation process. The effectiveness of the SSA-DRL framework is evaluated in sixteen different sections and is demonstrated through an ablation experiment and a comparison with the scheduled operation plan. By providing safe intelligent control of urban rail transit autonomous operation trains, this framework offers potential for practical implementation in the field.
Original article: https://arxiv.org/abs/2311.14457