MergeVLA：面向通用视觉-语言-动作智能体的跨技能模型合并
MergeVLA: Cross-Skill Model Merging Toward a Generalist Vision-Language-Action Agent
提出面向模型合并的VLA架构设计：用任务掩码稀疏激活LoRA适配器保持VLM骨干参数一致，并将动作专家中的自注意力替换为纯交叉注意力块以保持技能模块化，为多技能机器人策略合并提供可复用方案。
原文依据 MergeVLA introduces sparsely activated LoRA adapters via task masks to retain consistent parameters and reduce irreconcilable conflicts in the VLM.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期的视觉-语言-动作（VLA）模型通过数百万机器人演示对视觉-语言模型进行调优而重新构建。尽管它们在单一具身或任务族上微调时表现良好，但将其扩展到多技能场景仍具挑战性：直接合并在不同任务上训练的VLA专家会导致接近零的成功率。这引出一个根本问题：是什么阻止VLA在一个模型中掌握多种技能？通过对VLA微调过程中可学习参数的经验性分解，我们识别出两个不可合并性的关键来源：（1）微调使VLM骨干网络中的LoRA适配器朝着发散的、任务特定的方向漂移，超出了现有合并方法的统一能力。（2）动作专家通过自注意力反馈形成块间依赖，导致任务信息跨层扩散，阻碍模块化重组。为应对这些挑战，我们提出MergeVLA，一种在设计上保持可合并性的面向合并的VLA架构。MergeVLA通过任务掩码引入稀疏激活的LoRA适配器，以保持参数一致并减少VLM中不可调和的冲突。其动作专家将自注意力替换为纯交叉注意力块，使专业化保持局部化和可组合。当任务未知时，它使用测试时任务路由器从初始观察中自适应地选择适当的任务掩码和专家头，实现无监督任务推断。在LIBERO、LIBERO-Plus、RoboTwin以及真实SO101机械臂上的多任务实验中，MergeVLA取得了与单独微调专家相当甚至更优的性能，展示了跨任务、具身和环境的稳健泛化能力。项目页面：https://mergevla.github.io/
英文原文摘要 1819 字符
Recent Vision-Language-Action (VLA) models reformulate vision-language models by tuning them with millions of robotic demonstrations. While they perform well when fine-tuned for a single embodiment or task family, extending them to multi-skill settings remains challenging: directly merging VLA experts trained on different tasks results in near-zero success rates. This raises a fundamental question: what prevents VLAs from mastering multiple skills within one model? With an empirical decomposition of learnable parameters during VLA fine-tuning, we identify two key sources of non-mergeability:(1) Finetuning drives LoRA adapters in the VLM backbone toward divergent, task-specific directions beyond the capacity of existing merging methods to unify.(2) Action experts develop inter-block dependencies through self-attention feedback, causing task information to spread across layers and preventing modular recombination.To address these challenges, we present MergeVLA, a merging-oriented VLA architecture that preserves mergeability by design.MergeVLA introduces sparsely activated LoRA adapters via task masks to retain consistent parameters and reduce irreconcilable conflicts in the VLM.Its action expert replaces self-attention with cross-attention-only blocks to keep specialization localized and composable.When the task is unknown, it uses a test-time task router to adaptively select the appropriate task mask and expert head from the initial observation, enabling unsupervised task inference.Across LIBERO, LIBERO-Plus, RoboTwin, and multi-task experiments on the real SO101 robotic arm, MergeVLA achieves performance comparable to or even exceeding individually finetuned experts, demonstrating robust generalization across tasks, embodiments, and environments. Project page: https://mergevla.github.io/