Deep reinforcement learning with predictive auxiliary task for autonomous train collision avoidance

Antoine Plissonneau, Luca Jourdan, Damien Trentesaux, Lotfi Abdi, Mohamed Sallak, Abdelghani Bekrar, Benjamin Quost, Walter Schön

Abstract: Highlights•Deep reinforcement learning based method for autonomous train collision avoidance.•Auxiliary task of future state prediction.•The presented method has been compared to other policies and gives the best results.•Turning black-box agent to white-box agent for safety analysis.•Train collision avoidance RL-ready simulator.