The Gate, Not the Cache: Gate Provenance Bounds the Closed-Loop Reliability of Training-Free VLA Token Skipping

2026-07-31

AI Summary

This work addresses the reliability degradation of training-free vision-language-action (VLA) models in closed-loop control, which often stems from self-generated gating signals used for token skipping and can lead to catastrophic task failure. The study reveals that the instability arises not from the skipping mechanism itself but from the “noisiness” of these gating signals. To mitigate this, the authors propose an “actuation-relaxed refresh” strategy: during robot actuation, a single dense forward pass is unconditionally inserted to supply clean gating signals and refreshed key-value (KV) caches. This approach requires no additional triggers and integrates seamlessly into existing caching and pruning frameworks across diverse VLA architectures and tasks. Evaluated on LIBERO and SIMPLER benchmarks, the method fully recovers performance collapses caused by self-sampled gating, restoring closed-loop success rates to 0.98 while reducing service latency by 18–22% compared to dense inference, with consistent validation in both simulation and real-world robotic platforms.

Abstract

Token skipping is a widely used training-free way to accelerate vision--language--action (VLA) models by bypassing computation for most visual tokens at each control step according to a gate. When the next gate is harvested from the previous accelerated forward, however, the tokens skipped at one step are also the ones least visible to the next gate, and the damage can compound across control steps until the task fails. We study the two mechanisms this class is built on, reuse and deletion, crossing each against where its gate signal comes from on identical episodes. At a skip ratio of 0.9 on LIBERO-Object, both collapse when the gate comes from the model's own accelerated forwards, to 0.68 under reuse and to 0.31 under deletion against a dense 1.00, and the collapse is invisible to the action-level detectors we evaluate. What separates collapse from dense-level operation is not the mechanism but whether the gate is clean, computed by a forward that skipped nothing. We therefore propose actuation-slack refresh, one dense pass run while the robot executes its current action chunk, off the critical path, that hands the next step a clean gate and a fresh KV base. Since the measured detectors do not reliably reveal the failure, the refresh is unconditional rather than triggered. Both mechanisms then recover to 0.98, keeping the speed of skipping and the information of a dense pass. We then integrate the refresh into state-of-the-art caching and pruning methods across two VLA policies, 4 LIBERO suites, and 4 SIMPLER tasks, where it repairs every collapse caused by using a self-harvested gate. Serve latency drops 18--22\% below dense, measured both in simulation and on a physical robot. Where the gate signal comes from, not how tokens are skipped, decides closed-loop reliability for accelerated VLAs.

Problem

• token skipping
• gate provenance
• closed-loop reliability
• vision-language-action models
• training-free acceleration

Innovation

• token skipping
• gate provenance
• actuation-slack refresh
• closed-loop reliability
• training-free acceleration