HARVEE: research on AI-based hybrid system generating cooperative behavior between human and abstract object

Abstract

In this study, we aimed to realize cooperative interaction between a human and a physical agent. Based on our previous research, we developed the physical agent system HARVEE. The agent uses a physical body composed of an active pendulum mechanism driven by two servo motors and an internal spring structure to respond to human movements in real time, thereby generating a cooperative feedback loop between the human and the agent. A key feature of the system’s software algorithm is its ability to generate diverse and non-predefined behaviors by determining characteristics of human movement from skeleton data detected by a recognition AI, and smoothly fusing multiple motion modes according to the person’s movements through weighting. An experiment involving seven participants demonstrated that the system, by continuously changing its diverse motions, including both autonomous and tracking responses, was able to elicit sustained and exploratory behavior from participants, leading to the establishment of cooperative interaction even with minimal agent movement.