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

Social robots are increasingly used in a variety of service industries and are likely to integrate into human society in the foreseeable future¹. Unlike industrial robots, service robots frequently interact with humans². A core requirement for a service robot is thus the ability to navigate within a social setting in a human-like manner³. Human locomotion is remarkably influenced by social contexts⁴. For example, humans maintain comfortable distances from each other to avoid physical and psychological invasion of another’s personal space⁵. In addition, when two people are talking face-to-face, humans take a longer detour to avoid interrupting the interacting humans, regardless of the potential extra costs in time and energy⁶. Moreover, humans anticipate the intentions of others to prevent collisions with moving pedestrians⁷. These activities involve complex cognitive processing of social information⁸. Humans navigate social scenes based on social rules learned from extensive social life experiences⁹. Understanding these rules and applying them to the development of robots that can navigate in a socially aware manner is essential for the widespread use of service robots^10,11.

Two technical approaches have been used for embedding social rules into robotic algorithms. First, machine learning has been used to train neural networks based on a large sample of human navigation behaviours^12–14. Although this is an effective approach, how neural networks work remains mostly unclear, leading to the black box problem of artificial intelligence¹⁵. Moreover, algorithms based on an artificial neural network face the challenges of overfitting and lack of generality¹⁶. The second approach is to build up models based on theories from social science research. While the social sciences have extensively studied social rules, most of these studies identify social rules through natural observations and describe the rules on a conceptual level^17,18. A few studies borrow concepts from social sciences and use these concepts to design corresponding algorithms. For example, computer scientists used the concept of F-formation to successfully design algorithms for detecting social interactions in crowds¹⁹. Ishiguro and colleagues developed interactive humanoid robots that generated human-like behaviours through experimental research from the viewpoint of cognitive science and developed the Robovie...