Emotion Regulation Based on Multi-objective Weighted Reinforcement Learning for Human-robot Interaction

Given emotion is important in maintaining mental and physical well-being. A multi-objective weighted reinforcement learning (MOW-RL) decision method is proposed to execute emotion regulation in human-robot interaction. The goal of emotion regulation is to minimize the cost of executing the service while eliminate user’s negative emotions or maintain user’s positive emotions. Considering the coordination problem of two objectives, fuzzy analytic hierarchy process (FAHP) is used to calculate the weight of each target reward and punishment function under different conditions. In addition, the influence factors of different personality on the difficulty level of emotion transfer are calculated by FAHP. Experiments are performed by 20 experimenters in the laboratory scenario, from which the results show that experimenters’ satisfaction is 2.3, which is close to satisfaction.