Waist rotation angle as indicator of probable human collision-avoidance direction for autonomous mobile robots

Abstract

The likelihood of pedestrians encountering autonomous mobile robots (AMRs) in smart cities is steadily increasing. While previous studies have explored human-to-human collision avoidance, the behavior of humans avoiding AMRs in direct, head-on scenarios remains underexplored. To address this gap, we conducted a psychophysical experiment to observe how humans react to an AMR approaching directly. The AMR was programmed to approach from various starting points, including a direct path toward participants, and their avoidance movements were recorded. Participants were instructed to evade by moving either right or left, with no strong preference for a particular direction observed. This suggests that avoidance direction is not strictly influenced by individual factors, such as adherence to regional traffic norms. Additionally, motion analysis revealed that participants instinctively twisted their waists in the direction of avoidance before evading. Further experiments assessed the role of waist rotation angle in influencing human comfort during AMR avoidance. The results indicated that early AMR avoidance improved participant comfort. Moreover, using an RGB camera allowed non-contact measuring without sensors, broadening the applicability of the technique. These findings suggest that waist rotation reliably predicts avoidance direction, and non-contact detection methods, such as RGB cameras, show substantial potential for broader applications. Further research will focus on improving the accuracy and robustness of these non-contact techniques.