A Hybrid Human Motion Prediction Approach for Human-Robot Collaboration

Abstract

Prediction of human motion is useful for a robot to collaborate with a human partner. In this paper, we propose a hybrid approach for the robot to predict the human partner’s motion by using position and haptic information. First, a computational model is established to describe the change of the human partner’s motion, which is fitted by using the historical human motion data. The output of this model is used as the robot’s reference position in an impedance control model. Then, this reference position is modified by minimizing the interaction force between the human and robot, which indicates the discrepancy between the predicted motion and the real one. The combination of the prediction using a computational model and modification using the haptic feedback enables the robot to actively collaborate with the human partner. Simulation results show that the proposed hybrid approach outperforms impedance control, model-based prediction only and haptic feedback only.