Learning to Assist: Collaborative VLAs for Implicit Human-Robot Collaboration
For human-robot collaboration researchers, this paper identifies and mitigates a key failure mode in VLA-based collaborative systems, enabling longer-horizon tasks.
This work demonstrates that vision-language-action (VLA) models trained with imitation learning can support human-robot collaboration, but suffer from premature assistive actions due to action-chunking. They propose an inference-time steering method that reduces failures by 40% and improves task completion time by 25% in a user study.
Human-robot collaboration (HRC) combines the complementary strengths of humans and robots to improve task efficiency. However, many existing collaborative systems rely on hand-engineered pipelines, limiting their scalability and flexibility for new tasks. In this work, we show that models trained end-to-end with imitation learning, specifically vision-language-action (VLA) models, can support collaborative manipulation, and characterize the key factors affecting their real-world performance. We evaluate two state-of-the-art models and identify a failure mode of action-chunking policies in implicit HRC, where demonstration action leakage (i.e., action chunks crossing latent task transitions) can cause premature assistive behavior. We find that this issue increases with longer execution horizons and occurs in real-world collaborative VLA systems, such as when a robot attempts to hand over a tool before the person is ready. We propose an inference-time steering method to mitigate these erroneous assistive actions while preserving policy performance. Finally, through a 16-participant user study on a long-horizon collaborative assembly task, we show that steering enables a longer execution horizon while mitigating premature assistance, leading to faster collaboration and fewer failures compared to a shorter-horizon policy.