Article

Coupling of Execution Horizon and Sampling Depth in Flow-Matching VLA Policies

This paper investigates the coupling between numerical solver depth and execution horizon in flow-matching vision-language-action (VLA) models, finding that coarse-grained numerical integration reduces accuracy under rapid replanning but instead improves task performance under extended open-loop execution.

Part of the event: Coupling Between Execution Horizon and Sampling Depth in Flow-Matching VLA

This page provides a summary and source link only.