Three-Stage Training Framework for Multi-Task Sequential Human-Robot Collaboration

Research on sequential human-robot collaboration (HRC) has garnered significant attention. The goal is to maximize the synergy between humans and robots across sequential cooperations. However, limited research studies the sequential HRC problem across different tasks and current experimental environments are not able to support conducting such research. We introduce a three-stage training framework for multi-task sequential HRC. This framework characterizes human mental through their trust in robots and formulates cooperative robot policies. With the implementation of lifelong learning methods in the second stage, our framework overcomes the catastrophic forgetting problem successfully. We create a virtual simulation environment to support the study of multi-task sequential HRC research. Through quantitative and qualitative experiments in this environment, we validate our proposed framework and provide valuable insights for future research in sequential human-robot collaboration.