Motion Simulation of Flying Quadruped Robot Based on Deep Reinforcement Learning

摘要

Traditional motion control methods for fixed-profile aircraft are difficult to satisfy the requirements of flying quadruped robot. However, the emergence of artificial intelligence provides new ideas for the design of variant aircraft systems. During the deformed flight, the flying quadruped robot is extremely susceptible to various internal and external disturbances. Therefore, it is challenging to ensure its stability. On this basis, the motion problem of flying quadruped robot is investigated by employing the new deep reinforcement learning technology. The twin delayed deep deterministic policy gradient algorithm (TD3) is adopted to acquire gait and flight strategy through interactive learning with the environment. In order to enhance the learning efficiency in the training process, the framework of actor-critic learning system is proposed. In the meanwhile, the extra domain-specific knowledge is applied to shape the reward function. In addition, simulation studies are also carried out to verify the performance of the proposed TD3-based controller.