Learning Continuous Control through Proximal Policy Optimization for Mobile Robot Navigation

CORC > 沈阳自动化研究所 > 中国科学院沈阳自动化研究所 > 机器人学研究室

	Learning Continuous Control through Proximal Policy Optimization for Mobile Robot Navigation
	Zeng TP(曾太平)1,2,3
	2018
会议日期	December 7-8, 2018
会议地点	Hangzhou, China
关键词	Mobile Robots Deep Reinforcement Learning Continuous Control Proximal Policy Optimization Robot Navigation Mobile Robot Learning
页码	175-184
英文摘要	An intelligent mobile robot must be able to autonomously navigate in complex environments, so that it could be deployed in the real world. Traditional methods solve this problem by building a map of an environment, locating the position of the robot, and performing path planning to navigate the robot on the map. However, these methods often make a variety of assumptions and require intensive computational resources, which may restrict the application of these methods. More importantly, these methods lack of mechanisms to learn from failures. In this paper, I present a learning-based mapless mobile robot navigation method with continuous state and action spaces, in which a proved efficient policy gradient method, i.e. Proximal Policy Optimization (PPO), is introduced for learning continuous control tasks. It takes the normalized laser scanning data as input and directly outputs the continuous velocity commands to direct a mobile robot operating in the environments. The proposed method is trained end-to- end in several simulation environments to evaluate the performance without any manually designed features, human-provided labels, or prior assumptions. Experimental results show that it can learn to navigate through multiple different environments with a few hours of fully autonomous training. Also, it successfully learned to provide continuous control commands for mobile robots. Moreover, evaluations in multiple complex environments demonstrate the robustness and adaptability of the proposed method. The proposed learning-based method and mobile robot learning system can be a general approach to train mobile robots for more complex continuous tasks. Videos of the experiments can be found at https://youtu.be/P0bwzXI4EEA
产权排序	1
会议录	2018 International Conference on Future Technology and Disruptive Innovation
语种	英语
内容类型	会议论文
源URL	[http://ir.sia.cn/handle/173321/23862]
专题	沈阳自动化研究所_机器人学研究室
通讯作者	Zeng TP(曾太平)
作者单位	1.State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China 2.University of Chinese Academy of Sciences, Beijing 100049, China 3.Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang 110016, China
推荐引用方式 GB/T 7714	Zeng TP. Learning Continuous Control through Proximal Policy Optimization for Mobile Robot Navigation[C]. 见:. Hangzhou, China. December 7-8, 2018.