Researchers introduce UA-TOM, a lightweight belief-tracking module that detects behavioral regime changes in collaborative robot manipulation tasks. The method reduces post-switch collisions by 52% and achieves 85.7% detection accuracy by augmenting frozen vision-language-action models with state-space dynamics and causal attention.
Safety
Belief Dynamics for Detecting Behavioral Shifts in Safe Collaborative Manipulation
UA-TOM belief-tracking module cuts post-switch collisions by 52% in human-robot collaboration by detecting behavioral regime changes with 85.7% accuracy.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.LG (Machine Learning)BY sys://pipeline
Tags
safety