用模型自省判断何时重用视觉缓存，提升机器人控制精度。
Neural Introspection Gating for Adaptive KV-Cache Reuse in Vision-Language-Action Models
- 通过检测动作预测置信度下降来决定是否重用缓存
- 在静态画面下保持80%算力节省，恢复超100%准确率损失
- 无需训练、零成本添加，适合实时机器人系统
视觉-语言-动作（VLA）模型通过单一自回归Transformer将图像和语言指令直接映射为运动命令。在实时控制中，仍需大量计算重复重构几乎不变的视觉标记的键值（KV）表示。现有方法如VLA-Cache通过重用视觉静态区域的KV状态降低开销，但其策略仅依赖观测空间启发式，未考虑模型自身不确定性。本文提出Gated VLA-Cache，一种轻量级、无需训练的扩展，在视觉相似性缓存基础上引入神经自省机制。该方法在解码时监测前两个动作预测之间的对数概率差（置信度信号），当差值低于阈值时，触发缓存失效并重新计算。在四个LIBERO基准测试集上评估，使用OpenVLA与OpenVLA-OFT模型，当盲目缓存导致性能下降时，Gated VLA-Cache在LIBERO-Goal和LIBERO-Long任务中恢复超过100%的准确率损失，同时保留80%的算力节省。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。