Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (4): 17-30.DOI: 10.3778/j.issn.1002-8331.1810-0007

Previous Articles     Next Articles

Review of Imitation Learning Methods and Its Application in Robotics

LI Shuailong1,2,3, ZHANG Huiwen1,2,3, ZHOU Weijia1,2   

  1. 1.State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China
    2.Institute for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang 110016, China
    3.University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2019-02-15 Published:2019-02-19



  1. 1.中国科学院沈阳自动化研究所 机器人学国家重点实验室,沈阳 110016
    2.中国科学院 机器人与智能制造创新研究院,沈阳 110016
    3.中国科学院大学,北京 100049

Abstract: Imitation Learning(IL) has always been a research hotspot in the field of artificial intelligence. Imitation learning is a kind of method that reconstructs desired policies based on expert demonstrations. This method has been combined with methods such as reinforcement learning in theoretical research, and has achieved important results. In practical applications, especially in the complex environment of robots and other agents, imitation learning has achieved good results. This paper elaborates the research and application of imitation learning about robotics. Firstly, theoretical knowledge related to imitation learning is introduced. Secondly, two main methods of imitation learning are studied:Behavioral Cloning(BC) method and Inverse Reinforcement Learning(IRL) method. Thirdly, the successful applications of imitation learning are summarized. Finally, it is necessary to summarize the current problems and challenges and look forward to the future development trend.

Key words: artificial intelligence, behavioral cloning, inverse reinforcement learning, imitation learning

摘要: 模仿学习一直是人工智能领域的研究热点。模仿学习是一种基于专家示教重建期望策略的方法。近年来,在理论研究中,此方法和强化学习等方法结合,已经取得了重要成果;在实际应用中,尤其是在机器人和其他智能体的复杂环境中,模仿学习取得了很好的效果。主要阐述了模仿学习在机器人学领域的研究与运用。介绍了和模仿学习相关的理论知识;研究了模仿学习的两类主要方法:行为克隆学习方法和逆强化学习方法;对模仿学习的成功应用进行总结;最后,给出当前面对的问题和挑战并且展望未来发展趋势。

关键词: 人工智能, 行为克隆, 逆强化学习, 模仿学习