通过复用稳定动作和筛选视觉令牌，显著降低视觉语言动作模型推理开销。
Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models
- 识别动作连续性和视觉令牌冗余，提出免训练的加速机制
- 在LIBERO上降低55.7%计算量、36.0%延迟，成功率仅降0.7%
- 无需重训练，适合实时部署与边缘设备使用
视觉-语言-动作（VLA）模型通过自然语言指令实现通用机器人控制，但其高推理成本——源于大规模令牌计算与自回归解码——严重制约实时部署与边缘应用。现有工作多聚焦于架构优化，我们另辟蹊径，发现VLA模型存在双重冗余：(i) 相邻动作步骤间高度相似，(ii) 视觉令牌存在大量冗余。基于此，我们提出FlashVLA，首个无需训练、即插即用的加速框架，支持动作复用。FlashVLA通过感知令牌的动作复用机制避免稳定动作步骤间的冗余解码，并采用信息引导的视觉令牌选择策略剔除低贡献令牌。在LIBERO基准上的实验表明，FlashVLA将FLOPs减少55.7%，延迟降低36.0%，任务成功率仅下降0.7%。结果证明其可在不重新训练的前提下实现轻量化、低延迟的VLA推理。
原文摘要 · Abstract (English)
Vision-Language-Action (VLA) models have emerged as a powerful paradigm for general-purpose robot control through natural language instructions. However, their high inference cost-stemming from large-scale token computation and autoregressive decoding-poses significant challenges for real-time deployment and edge applications. While prior work has primarily focused on architectural optimization, we take a different perspective by identifying a dual form of redundancy in VLA models: (i) high similarity across consecutive action steps, and (ii) substantial redundancy in visual tokens. Motivated by these observations, we propose FlashVLA, the first training-free and plug-and-play acceleration framework that enables action reuse in VLA models. FlashVLA improves inference efficiency through a token-aware action reuse mechanism that avoids redundant decoding across stable action steps, and an information-guided visual token selection strategy that prunes low-contribution tokens. Extensive experiments on the LIBERO benchmark show that FlashVLA reduces FLOPs by 55.7% and latency by 36.0%, with only a 0.7% drop in task success rate. These results demonstrate the effectiveness of FlashVLA in enabling lightweight, low-latency VLA inference without retraining.
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。