VLA Knows Its Limits

TL;DR

This paper investigates the impact of execution horizon in flow-based Vision-Language-Action models, revealing performance fluctuations and proposing a dynamic estimation method, AutoHorizon, to adapt to environmental changes.

Contribution

It introduces AutoHorizon, the first method to dynamically estimate execution horizon at test time, improving adaptability and performance in flow-based VLA models.

Findings

Performance varies with execution horizon, initially improving then declining.

AutoHorizon effectively adapts horizon, enhancing robotic manipulation tasks.

AutoHorizon generalizes across tasks and models with minimal overhead.

Abstract

Action chunking has recently emerged as a standard practice in flow-based Vision-Language-Action (VLA) models. However, the effect and choice of the execution horizon - the number of actions to be executed from each predicted chunk - remains underexplored. In this work, we first show that varying the execution horizon leads to substantial performance deviations, with performance initially improving and then declining as the horizon increases. To uncover the reasons, we analyze the cross- and self-attention weights in flow-based VLAs and reveal two key phenomena: (i) intra-chunk actions attend invariantly to vision-language tokens, limiting adaptability to environmental changes; and (ii) the initial and terminal action tokens serve as stable anchors, forming latent centers around which intermediate actions are organized. Motivated by these insights, we interpret action self-attention weights as a proxy for the model's predictive limit and propose AutoHorizon, the first test-time method that dynamically estimates the execution horizon for each predicted action chunk to adapt to changing perceptual conditions. Across simulated and real-world robotic manipulation tasks, AutoHorizon is performant, incurs negligible computational overhead, and generalizes across diverse tasks and flow-based models.