Route-Guided CBF-QP Repair for Safe Vision-Language-Action Models
What it is
This paper proposes a post-hoc repair module for Vision-Language-Action (VLA) models that uses predicted action chunks to detect blocked trajectories. It generates entry-exit-rejoin guidance around obstacles and adds this route as a soft term in a Control Barrier Function (CBF) Quadratic Program (QP) while keeping safety constraints hard.
The problem it solves
Standard CBF-QP safety filters can push robots away from learned task trajectories, leaving VLA policies without a reliable recovery direction when obstacles block the nominal path.
What you could use it for
- Improving collision avoidance in robot manipulation tasks
- Enhancing task success rates for VLA models in obstacle-rich environments
- Providing recovery directions for VLA policies when nominal trajectories are blocked
Why it matters
Enables Vision-Language-Action models to operate safely in complex physical environments by reconciling task completion with strict safety constraints, potentially increasing the reliability of autonomous robots.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.