In order to solve the traffic problem caused by the popularization of cars, autonomous driving technology has developed rapidly. Reinforcement learning is an effective way to solve the decision-making problems of autonomous driving. However, the decision method based on RL has many tacit assumptions, which seriously limits the possibility of application in the actual driving environment. Therefore, this paper proposes a novel reinforcement learning architecture to solve these limitations. Firstly, we use bootstrapped DQN to make decisions that have long-term advantages. Secondly, a simulation environment based on vehicle dynamics model is established to obtain more realistic training data. Furthermore, the artificial potential field method is added to the reward function to avoid collisions more effectively, thereby improving the vehicle safety. Finally, the simulation results show that the proposed reinforcement learning scheme is effective to guarantee the vehicle safety in typical driving conditions.
A Bootstrapped DQN-Based Decision-Making Method for Autonomous Driving Vehicles
2023-10-16
2486031 byte
Conference paper
Electronic Resource
English
Ethical decision making for autonomous vehicles
IEEE | 2020
|Autonomous Driving Cars: Decision-Making
Springer Verlag | 2020
|ETHICAL DECISION MAKING FOR AUTONOMOUS VEHICLES
British Library Conference Proceedings | 2020
|Autonomous Vehicles Implementing User-Based Ethical Decision Making
European Patent Office | 2023
|