Unmanned driving test scene automatic generation method based on reinforcement learning

• Main Authors: REN ZHILEI, WANG JINGBO, WANG HAIBO, ZOU PEIYU, ZHOU ZHIDE, JIANG HE
• Format: Patent
• Language: Chinese; English
• Published: 01.09.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention belongs to the technical field of unmanned driving test, and particularly relates to an unmanned driving test scene automatic generation method based on reinforcement learning. According to the method, the dynamic elements in the scene are regarded as the intelligent agents, the model is continuously trained in the simulation process through reinforcement learning, the intelligent agents learn to interact with the self-vehicle in the driving process, and therefore the unmanned vehicle faces the more complex scene. Due to the fact that a plurality of pedestrians and a plurality of vehicles exist in a scene, a centralized training decentralized execution framework and an actor-commentator thought are adopted to enable all agents to cooperate. In the training stage, each dynamic element needs to inform a central processing unit module of the observed value of the dynamic element and the current action, and the action which should be made by each agent in the next step is determined through the info