GeoAAC Adapts Action Horizons in VLA Policies Using Flow Geometry
September 18, 2026
GeoAAC uses the geometry of Flow Matching denoising trajectories to adapt action horizons in Vision-Language-Action policies. By correlating geometric variation with predictive uncertainty, the method adjusts action chunking based on real-time prediction reliability.
HOW THIS AFFECTS YOU
●
builderThis enables more precise control in robotics by adjusting execution frequency based on model confidence.
●
researcherThis method provides a way to use denoising trajectory geometry as a proxy for uncertainty.