Decision-Making for Complex Scenario using Safe Reinforcement Learning

Abstract

In recent years, machine learning is widely used in many fields. Compared with the rule-based method, machine learning plays a more excellent role in the decision-making of the autonomous vehicle. Some complex situations are often met in our daily life. To this end, Safe reinforcement learning(RL) is introduced to ensure that safer actions are selected. Constant Turn Rate and Acceleration(CTRA) model is first used to predict the future trajectories of surrounding vehicles. Then Double Deep Q-Learning(DDQN) method is used to make decisions and ensure the autonomous vehicle can move at the desired speed as much as possible. In order to achieve a safer decision-making, some safety rules are introduced. Finally, the algorithm is demonstrated in Simulation of Urban Mobility(SUMO) and has been proved to have an outstanding performance on such a complex scenario.