Robot trajectory tracking control using learning from demonstration method

摘要

This paper addresses robot trajectory tracking problem by using the learning from demonstration (LFD) method. Firstly, the trajectory tracking problem is formulated and the related previous works are introduced. Secondly, a trajectory tracking control policy using a three-layer neural network method, i.e., extreme learning machines (ELM), is proposed to minimize the real-time position and velocity errors. In the proposed method, the control algorithms are learnt from demonstrations directly such that the parameter adjusting problem in the traditional model-based methods is avoided. Besides, the trained controller has generalization ability to unseen situations which can be used to track different desired trajectories without any extra re-training. Thirdly, the stability analysis of the proposed control algorithm is provided and the corresponding parameter constraints are derived. Finally, the effectiveness and the generalization ability of the proposed control algorithms are demonstrated and discussed with simulation and experimental examples. (C) 2019 Elsevier B.V. All rights reserved.