Variable Gesture-based Haptic Guidance for Human-Robot Interaction

This paper presents a gesture-based variable impedance control approach for haptic guidance in humanrobot interaction. In this approach, applied forces have adaptively been calculated in real time with respect to gestural differences between user motions and reference models. A virtual task has been segmented into primitive gestures and reference hidden Markov models (HMMs) have been trained according to these gestures. Our approach is partially task independent and robust to spatial variation of gestures due to the HMM-based segmentation scheme. Our experimental results show that the variable impedance control can effectively improve user performance in the presence of human behavior uncertainty during a collaborative task. This model-based haptic guidance could be extended to provide haptic guidance in real-time skill assessment, sport training, and rehabilitation exercises.