Neural Introspection Gating for Adaptive KV-Cache Reuse in Vision-Language-Action Models

arXiv:2608.10824v1 Announce Type: cross Abstract: Vision-Language-Action(VLA) models map camera images and language instructions directly to motor commands through a single autoregressive transformer. In real-time control, they still spend substantial compute recomputing key-value(KV) representations for visual tokens that barely change across neighboring frames. Recent work such as VLA-Cache reduces that cost by reusing KV states for visually static patches, but its policy relies only on observation-space heuristics and does not account for the model's own uncertainty. We propose Gated VLA-Cache, a lightweight, training-free extension that augments visual-similarity caching with neural introspection. The method monitors the logit margin between the top two predicted action tokens, a zero-cost confidence signal available during decoding. When the margin drops below a threshold, the cache is invalidated and a full recompute is triggered. Evaluated on four LIBERO benchmark suites with both OpenVLA and OpenVLA-OFT, Gated VLA-Cache improves reliability when blind caching hurts. On LIBERO-Goal and LIBERO-Long, it recovers over 100% of the lost accuracy while retaining 80% of the compute savings.

Related

• CrossVLA: Cross-Paradigm Post-Training and Inference Optimization for Vision-Language-Action Models
• Pixels for Programs? A Cross-Provider Case Study of Input-Token Accounting for Source Code as Text and Images
• SlimVLM: Sensitivity-aware Dynamic Structured Pruning with Adaptive Visual Token Selection for Efficient Vision-Language Models
• BLUE: Toward Better Language Use in Efficient Vision-Language-Action Models for Autonomous Driving

Source: arXiv cs.CV | 2026-08-12