Manipulator Meta-Imitation Learning Algorithm with Memory Weight Integration

2019 2nd International Conference on Advanced Materials, Intelligent Manufacturing and Automation

Yin, Mingjun^1 ; Zeng, Qingshan^1

School of Electrical Engineering, Zhengzhou University, Zhengzhou

450001, China^1

关键词: Capacity modeling;    Imitation learning;    Key characteristics;    Multiple tasks;    Multitask learning;    Process of learning;    Unknown environments;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/569/5/052039/pdf
DOI  :  10.1088/1757-899X/569/5/052039

来源: IOP

【 摘 要 】

Versatility is one of the key characteristics of general agent. In order to enable the manipulator to quickly and effectively acquire the ability to perform multiple tasks in an unknown environment, a large capacity model is essential. In this paper, the memory weight integration term adapted to meta-learning algorithm is proposed. By adjusting the plasticity of neurons, the manipulator can learn to learn more effectively in the process of learning multi-task and improve the forgetting problem of multi-task learning. Then, this paper combines the memory weight integration with meta-imitation learning, so that the manipulator can acquire new skills from a single demonstration task. Finally, a 7-DoF manipulator in PusherEnv experiment is used to explore the influence of different integration coefficients on the algorithm. The results show that the memory weight integration can effectively improve the success rate of tasks.