Human Motion-relied Social Interaction Intention Prediction for Autonomous Mobile Robots

To behave safely and socially in human-robot shared workplaces, autonomous mobile robots (AMRs) should not only be able to detect and track people’s motions but also predict their social interaction intention (SII). In this study, we address the problem of human motion-relied social interaction intention prediction using an onboard sensor-based multi-people detection and tracking approach. In the first step, we present an integration of human detection and tracking through combining LiDAR-based leg detector and RGB-D-based YOLO human detection supported by social characteristics-oriented scan-to-track data association. A social dynamic confidence function is generated to analyze the reliability of each person tracked using the leg detector using human social norms and dynamic features, then combined with the YOLO-based confidence factor to increase the reliability of the tracking system. In the second step, we address social interaction intention prediction for individuals and groups of people using human relative distance, relative motion orientation, and social characteristics. To examine and validate the proposed methods, two sets of real-world experiments of multi-people tracking and social interaction intention prediction were designed, resulted in 95% tracking accuracy and 74% intention estimation accuracy in challenging social situations.