Event
Proposes FIT-MAC Closed-Loop Multi-Agent VLA Robustness Framework
1 reports · 1 sources
Jiatong Li et al. proposed FIT-MAC on arXiv, a failure-aware closed-loop multi-agent control framework that adds a control layer at inference time to visual-language-action (VLA) policies to improve robustness. When no intervention is needed, the pretrained VLA remains the default action generator. The paper states that experiments on the LIBERO-PRO benchmark show FIT-MAC improves robustness across multiple VLA backbone models.
Report timeline
- FIT-MAC: A Closed-Loop Multi-Agent Framework for Improving the Robustness of Vision-Language-Action Models
The paper proposes FIT-MAC, a failure-aware closed-loop multi-agent control framework that adds a control layer at inference time to vision-language-action (VLA) policies to improve robustness.
arXiv cs.RO ·