Co-GAIL: Learning Diverse Strategies for Human-Robot Collaboration

Advancing technologies for human-robot collaboration (HRC) has the potential to unlock applications with large societal impact in manufacturing, hospitals, and home settings [1, 2]. However, robots that are designed to work around humans are still limited in versatility when performing collaborative tasks. Recent advances in robot learning focus on robots that work in isolation [3, 4, 5, 6] or alongside other agents that do not exhibit human traits [7, 8, 9, 10]. Learning to collaborate with humans presents unique challenges to existing robot learning methods: instead of optimizing only for efficient task completion, the robot assistant must act in coordination and adapt to the diversity of strategies and movements of their human counterparts. This work aims to develop robot assistants that adapt to diverse human strategies and movements in collaborative manipulation tasks.