Multi-robot behavior adaptation to local and global communication atmosphere in humans-robots interaction

Multi-robot behavior adaptation mechanism based on cooperative–neutral–competitive fuzzy-Q learning is proposed for coordinating local communication atmospheres in humans-robots interaction, in which the communication atmosphere is represented by a two-layer fuzzy fusion model and is visualized by shape–color–fill–wave graphics. It aims to realize smooth communication between humans and robots in the local and global communication atmosphere coexistent interaction by decreasing the response time of robots and social distance between humans and robots, as well as visualizing the communication atmosphere. Experiments on multi-robot behavior adaptation are performed in a virtual home party environment. Results show that the proposal saves 47 and 103 learning steps (i.e., the learning rate is increased by 72 % and 85 %) compared to fuzzy production rule based friend-Q learning (FPRFQ) and friend-Q learning (FQ), respectively; the distance between human-generated atmosphere and robot-generated atmosphere is 3 times and 7 times shorter than the FPRFQ and the FQ, respectively. Additionally, subjective estimation of graphic visualization of the atmosphere through questionnaire obtains 85.5 % accuracy for shape, 77.2 % for color, 65.3 % for fill, and 91.7 % for wave. The proposed mechanism is being extended to the robot behavior adaptation to international communication atmosphere, where the atmosphere is generated by people from different countries with different cultural backgrounds.