基于深度强化学习的四足机器人研究综述

Review of quadruped robot research based on deep reinforcement learning

随着机器人和人工智能技术的快速发展,行业对智能机器人的需求提高,提升机器人智能化水平是急需解决的问题。四足机器人作为具有代表性的腿足式机器人之一,因其友好的外形和结构优势,广受研究人员喜爱。首先介绍了目前国内外具有代表性的研究团队开发出的几款较为先进的四足机器人平台。然后介绍了一些研究人员使用深度强化学习方法训练四足机器人进行自主学习运动的工作。并分析了更适合进行基于深度强化学习方法训练四足机器人控制策略的仿真平台及环境,总结目前由仿真环境到四足机器人实体平台存在的难点及减小仿真到现实之间差距的思路。最后,对具有学习能力的四足机器人的应用场景进行展望,并介绍了未来研究工作的思路。

With the rapid development of robots and artificial intelligence technology,the demand for intelligent robots in many fields is increasing,and improving the level of robot intelligence is an urgent problem to be solved.As one of the representative legged robots,the quadruped robot is widely liked by researchers because of its friendly appearance and structural advantages.This paper first introduces several advanced quadruped robot platforms developed by representative research teams at home and abroad.Then the work of some researchers using deep reinforcement learning to train quadruped robots for autonomous learning is introduced.It also analyzes the simulation platform and environment that are more suitable for training quadruped robot control strategies based on the deep reinforcement learning method,summarizes the current difficulties from the simulation environment to the quadruped robot entity platform,and the ideas of reducing the gap between simulation and reality are discussed.Finally,the application scenarios of quadruped robots with learning ability are prospected,and the ideas for future research work are introduced.