Autonomous Emergency Steering Using Deep Reinforcement Learning For Advanced Driver Assistance System

Misako Yoshimura, Gakuyo Fujimoto, Abinav Kaushik, Bharat Kumar Padi, Matthew Dennison, Ishaan Sood, Kinsuk Sarkar, Abdul Muneer, Amit More, Masamitsu Tsuchiya, Satoru Araki, Anil Hebber, Tijmen Tieleman, Yuji Yasui

Abstract: Although automobile technology has continuously evolved in recent years, there are still many traffic accidents all over the world. In order to reduce the number of traffic accidents, we focused on developing an emergency collision avoidance system. In this work, our goal is to propose a smart controller which can calculate the trajectory in order to avoid vehicle collision in the emergency situation while preventing the potential secondary collision caused by the change of trajectory in order to avoid the first collision. We consider two emergency driving scenarios as a first step towards this goal and propose a solution using deep reinforcement learning algorithm. We present detailed results of training and validation in the simulated environment. Further, we also validate the proposed solution in an actual vehicle and show that the trained controller successfully avoided collisions in case of an emergency situation without any human interaction.