Human-behaviour-based social locomotion model improves the humanization of social robots

The widespread use of robots in service fields requires humanoid robots that mimic human social behaviour. Previous quantitative studies exist in human social behaviour, but engineering social robots requires translating these findings into algorithms to enable reliable and safe robot locomotion. To bridge this gap, we first quantitatively investigate the social rules that apply when people pass one another in social settings in laboratory and real-world experiments. We then developed a social locomotion model based on these observations to predict human path selections and walking trajectories in complex dynamic social scenes. The model was implemented into a socially aware navigation algorithm for a service robot. The robot navigating by the social locomotion algorithm behaved more like humans and received higher comfort ratings compared with previous social navigation algorithms tested. The model sheds new light on how to directly translate the findings of human behavioural experiments into robotic engineering.