Title:Robustifying Long-term Human-Robot Collaboration through a Hierarchical and Multimodal Framework

Abstract:Long-term Human-Robot Collaboration (HRC) is crucial for developing flexible manufacturing systems and for integrating companion robots into daily human environments over extended periods. However, sustaining such collaborations requires overcoming challenges such as accurately understanding human intentions, maintaining robustness in noisy and dynamic environments, and adapting to diverse user behaviors. This paper presents a novel multimodal and hierarchical framework to address these challenges, facilitating efficient and robust long-term HRC. In particular, the proposed multimodal framework integrates visual observations with speech commands, which enables intuitive, natural, and flexible interactions between humans and robots. Additionally, our hierarchical approach for human detection and intention prediction significantly enhances the system's robustness, allowing robots to better understand human behaviors. The proactive understanding enables robots to take timely and appropriate actions based on predicted human intentions. We deploy the proposed multimodal hierarchical framework to the KINOVA GEN3 robot and conduct extensive user studies on real-world long-term HRC experiments. The results demonstrate that our approach effectively improves the system efficiency, flexibility, and adaptability in long-term HRC, showcasing the framework's potential to significantly improve the way humans and robots work together.