外观
论文arXiv 2606.277554 个月前
VLA模型语言骨干冗余度高，DTR协议验证剪枝后性能反升
本文由 AI 根据公开内容整理，可能有误，请以原文为准：Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?（arXiv 2606.27755）
核心内容
针对VLA模型继承自预训练VLM的语言骨干容量过剩问题，研究者提出Drop-Then-Recovery（DTR）分析协议，通过受控移除Transformer块并微调来评估架构冗余度。实验显示语言模块对标准操控任务呈现强不对称冗余性，在LIBERO基准上，OpenVLA-OFT移除50% LLM块后成功率由95.0%提升至98.3%，且仅保留2个语言块仍能恢复基线性能，表明当前基准对深层语言 grounding 压力有限。
为确保剪枝干预可靠性，研究提出GateProbe单次虚拟门灵敏度指标，用于量化排序各模块对下游动作Loss的贡献度。跨架构与真机工业场景测试表明，相比语言骨干，视觉与动作通路对模块移除容忍度显著更低。该发现指出未来VLA设计应重新分配容量权重，避免盲目堆叠语言参数，相关代码已在GitHub开源供社区复现与优化。
关键看点
- DTR协议：通过移除Transformer块后微调测量容量必要性，证实语言骨干在机器人操控中存在高度结构性冗余
- 性能反升：OpenVLA-OFT在LIBERO基准移除半数LLM块后成功率从95.0%提升至98.3%，同等微调预算下表现更优
- GateProbe指标：提出单次虚拟门灵敏度度量，精准排序模块对下游动作Loss贡献，指导非均匀剪枝策略
- 模态不对称：视觉与动作通路对模块移除敏感度远高于语言骨干，仅保留2个语言块即可维持基线操控性能
工程与产业观察
该研究为具身大模型轻量化提供实证依据，证明当前VLA语言骨干存在显著过参数化。工程上可采用DTR协议与GateProbe指标进行结构化剪枝，在保持操控性能前提下大幅降低推理开销。同时警示行业需构建更具挑战性的语言 grounding 评测基准，避免模型在简单指令集上虚高。
英文摘要 (Abstract)
Vision-Language-Action (VLA) models enable instruction-driven robotic manipulation, but they inherit oversized language backbones from pretrained VLMs whose capacity far exceeds what is needed for short robotic instructions. This raises a basic question: how much of a VLA model is actually necessary for closed-loop control? In this work, we study architectural redundancy in VLA models by using transformer block removal as a controlled intervention. We introduce \textbf{Drop-Then-Recovery (DTR)}, an analysis protocol that removes selected blocks from a pretrained VLA model and then fine-tunes the resulting model to measure whether the removed capacity was necessary for downstream control. To make this intervention reliable, we propose \textbf{GateProbe}, a one-shot virtual-gate sensitivity metric that ranks blocks by their contribution to the downstream action loss. Across multiple VLA architectures, manipulation benchmarks and even real-robot industrial scenarios, we find a strong asymmetry in post-removal recoverability: \ul{\textit{language backbones are highly redundant for standard robotic manipulation tasks, whereas vision and action pathways are substantially less tolerant to removal}}. On LIBERO, removing half of the LLM blocks even improves OpenVLA-OFT from 95.0% to 98.3% under the same downstream fine-tuning budget, and retaining only two language blocks still recovers baseline-level performance. These results suggest that current VLA benchmarks may exert limited pressure on deep language grounding and compositional instruction understanding, and that future VLA architectures should allocate capacity more deliberately across language, vision, and action components. The code is available at https://github.com/s1ghhh/VLADrop.
作者：Guoheng Sun, Kaixi Feng, Shwai He, Xiaochuan Gong, Yexiao He, Ziyao Wang, Zheyu Shen, Wanghao Ye, Ramana Rao Kompella, Gaowen Liu, Ang Li
来源
本页事实对应的公开页面，以原文为准。