Adaptive Action Horizon Selection via Flow-Field Uncertainty

Reliable uncertainty estimates are important for adaptive execution in vision-language-action (VLA) models. Existing approaches such as Adaptive Action Chunking (AAC) estimate action entropy from multiple independently sampled action chunks, introducing additional inference cost and capturing both uncertainty within an action trajectory and diversity across alternative strategies. We introduce Divergence-based Uncertainty-aware Adaptive Action Chunking (DUAC), a post-hoc method that extracts trajectory-conditioned predictive uncertainty from the vector field of a pretrained flow-matching VLA. Building on a closed-form flow-posterior covariance formulation, DUAC derives chunk- and action-level uncertainty scores from one generated action chunk using Jacobian--vector products, without retraining, auxiliary confidence heads, or multiple endpoint samples. DUAC broadly agrees with AAC's sampling-based entropy while being less sensitive to diversity among distinct but valid action strategies and substantially reducing uncertainty-estimation latency. We use the resulting per-action uncertainty profile for adaptive horizon selection. Across LIBERO, LIBERO-Pro, LIBERO-Plus, and RoboCasa, DUAC remains competitive with entropy-based AAC and improves over fixed-horizon execution in several settings, including under distribution shift. These results establish the flow vector field as an efficient, trajectory-conditioned source of predictive uncertainty for robot decision making.