A Pomdp-Based Approach for Obstacle Avoidance in Autonomous Trains

Autonomous trains must operate in highly dynamic environments where ensuring safety remains a significant challenge.Unlike human operators who can intuitively assess and respond to potential risks, autonomous systems require continuous, real-time evaluation of their surroundings in order to make safe decisions.In this paper, we present an approach for obstacle avoidance and environment monitoring for autonomous trains using Partially Observable Markov Decision Processes (POMDPs).The proposed approach models and assesses the risks while take into account the uncertainties associated with the train status and the various operational and environmental conditions; then outputs the adequate control action to maintain the train in safe state.To evaluate its efficiency, the approach is applied to the anti-collision function of autonomous trains in hazardous scenarios.