Avoiding Chatters with Hesitation in Human–Robot Co-Learning

Efficient human–robot collaboration requires co-learning, which is a mutual adaptation process where both partners observe each other and adapt their behavior accordingly. The evolution of the joint co-learning system can demonstrate chatter, which is a specific limit cycle phenomenon where two learners both adjust their behaviors to correct for the mismatch, but cannot settle on a steady state. This paper provides rigorous modeling and analysis of co-learning between humans and robots from a control perspective. In this paper, the properties of chatter are quantitatively analyzed by modeling the human–robot co-learning system as a mutual feedback system, where the adaptation processes of humans and robots are characterized by different online learning strategies according to the inherent asymmetry of cognitive ability and transparency between them. Inspired by the hesitation phenomenon that has been commonly observed in human–human negotiation, a chatter-avoiding algorithm is then proposed to avoid chatter during the co-learning process, whose effectiveness is validated by theoretical proofs and experimental results.