Key points are not available for this paper at this time.
In this article, a learning-based controller is proposed to realize motion policy learning based on intuitive human demonstrations. The position, velocity, and force data during the demonstration are collected as input features without any physical contact with the human demonstrator, and an algorithm is designed to automatically label the data in combination with motion and force data. After the learning process, the robot can complete the assembly according to the human demonstrations, and the proposed controller will generate different angular acceleration commands as control inputs to help finish the manipulation well. Finally, a comprehensive analysis, including Lyapunov stability and Lipschitz constraint, is also provided to guarantee the stability and security of this learning-based controller. Sufficient experiments based on the real robot system verify the effectiveness of the proposed method.
Sheng et al. (Mon,) studied this question.