Kian Hosseinkhani (Simon Fraser University) et al. · 用IMLE替换扩散/流匹配动作头，实现单步动作生成，在保持视觉语言骨干跨任务泛化的同时大幅降低推理延迟。可直接替换现有VLA的动作头做加速对比。
Jianming Ma et al. · 为流匹配VLA与世界动作模型提供可微、且与训练过程对齐的硬约束施加机制，使生成动作满足物理可行性约束。可直接套到现有flow-matching动作头上做安全部署。
Chenyu Su et al. · 针对VLA在精密、可重复任务上的不可靠，提出非对称协同自举做高效真机在线RL后训练，让VLA在真实试错中自主提升精度。可直接用于真机RL精调流程。
Shiyuan Dong et al. · 用光纤布拉格光栅（FBG）触觉阵列配合自监督对比学习Transformer，从机器人触诊信号中估计肿瘤深度。触觉表征自监督思路可借鉴，但任务为医疗触诊，与VLA操作策略无直接衔接。
主张记忆不必放在动作策略内部，而是以智能体侧记忆引导可转向的动作模型完成长时程操作，并厘清VLA+规划器+几何工具系统的增益归因。系统集成思路值得读，摘要未给量化对比。
研究具身多智能体的组织形式如何决定集体智能，提出按任务动态改变机器人团队组织结构的方法，而非固定拓扑。属多机协作组织层面研究，未涉及VLA策略与操作实验细节。
Jinho Jeong et al. · arXiv:2609.10706v1 Announce Type: new Abstract: Human video datasets have emerged as a compelling alternative to expensive real-robot data, offering rich diversity at scale. To bridge the human-to-robot embodiment gap, existing approaches either robotize videos in task-matched settings or address observation and action alignment separately at scale. In this work, we systematically examine whether robotized human videos can provide effective and scalable supervision for pretraining vision-languag
Jiawen Wang et al. · 针对杂乱真实场景缺乏显式避障机制的问题，学习障碍感知表征来提升扩散策略的泛化与运动生成质量。属操作策略层面的表征改进，无VLA规模的数据或跨具身验证。
Chuhan Meng et al. · 用有限线性时序逻辑（LTL）约束引导数据生成并训练扩散策略，解决多智能体操作中的时序、逻辑、安全约束与智能体不同步问题。多机协作方向可参考，但非VLA主流路线。
Pengfei Zhang et al. · 指出已有RL只扰动生成策略的噪声空间而无法调控中间动作表征，提出双隐空间RL同时调制噪声与中间表征。对生成式VLA的RL精调有直接启发。
Haoran Pei et al. · 不做额外训练，直接读取冻结预训练机器人世界模型的内部预测状态作为在线失败检测信号，免去专用监控模块。对世界模型驱动的VLA部署监控有复用价值。
Wei Li et al. · arXiv:2609.11875v1 Announce Type: new Abstract: Recent advances in Vision-Language-Action (VLA) models have improved robotic manipulation, yet observation-to-action learning remains limited by a fundamental transition realizability gap, manifested in three tightly coupled problems: (i) Transition ambiguity. Visually similar current observations may correspond to different manipulation phases and imply different subsequent transitions. (ii) Prediction--execution mismatch. A visually plausible pre
Ning Gao et al. · 面向通用移动操作策略的仿真诊断基准，26个任务、5个能力维度、4个泛化轴，突破单一成功率标量评价。移动操作评测可直接借用，但属基准类工作不紧急。
Long Xu et al. · 用智能体自动生成可交互、功能合理的物体组合场景，为机器人操作基础模型提供可扩展的仿真评测与数据生成。仿真数据管线方向可参考。
Peidong Wang et al. · arXiv:2608.11350v2 Announce Type: replace-cross Abstract: Embodied agents are increasingly built as systems around foundation models, where performance depends not only on model weights but also on the skills, context, action interfaces, and execution harness surrounding the model. While supervised fine-tuning and reinforcement learning can adapt agents to new environments, they require additional data, rewards, and training runs; meanwhile, many train-free code-centric approaches rely on progra
Shengye Dong et al. · 把动作chunk视为短时多元轨迹，用时间-频率几何交叉注意力显式建模其频域结构，改进chunked VLA的动作表征。对chunk长度/平滑性调优有直接参考价值。
Yang Wan et al. · 系统梳理机器人策略验证器——从成功检测器、奖励模型到运行时监控器，及其用于VLA评测与训练的方式。属重要综述，作为查阅入口而非即时方法。
Yanzhe Chen et al. · arXiv:2609.10522v1 Announce Type: cross Abstract: Foundation vision-language models (VLMs) exhibit broad intelligence about the world, yet translating this intelligence into robot control remains challenging. We present Show-Harness, an Embodied Harness that enables VLMs to "play" robots through a compact semantic interface linking intent to action. Show-Harness exposes discrete semantic action units that VLMs can naturally reason over, while embodiment-specific interpreters deterministically gr
Haihao Lin et al. · arXiv:2606.08653v2 Announce Type: replace-cross Abstract: Action-supervised fine-tuning of vision-language-action (VLA) policies fits demonstrations effectively but constrains only the directions that change predicted actions, leaving visual structure consistent across action-equivalent states free to collapse. We formalize this as residual visual collapse along local action fibers and propose FiberTune, a training-time objective that preserves teacher-structured visual residuals without adding
Yudong Gao et al. · 首次对量化VLA实施比特翻转攻击：少量梯度筛选的翻转即可把闭环成功率打到0，并揭示动作解码结构决定脆弱性。对量化VLA部署鲁棒性有警示价值。
Tengbo Yu et al. · 让人类与机器人共享同一外骨骼，实现一对一灵巧演示采集，用闭环映射解决可穿戴外骨骼只能开环重定向、演示无法忠实迁移的问题。灵巧操作数据采集可参考。
Kai Stewart et al. · 用实时雅可比估计让拟人手快速学会在手内转笔书写，绕开大量建模与数据采集。属灵巧手控制方法，与VLA策略学习路径不同。
Hongzhe Bi et al. · 把感知、预测、动作、评估、改进统一进一个自演化世界模型，而非简单在世界模拟器上挂动作头，面向灵巧操作。世界模型+灵巧操作方向可跟踪，但需核实实验规模。
Qinzhen Ma (Rice University) · arXiv:2609.09597v2 Announce Type: replace Abstract: Accurate tactile forecasts need not improve force-constrained control. We study a 652,157-parameter action-conditioned visuotactile world model with matched behavior cloning, policy learning in imagination, independent reactive implicit Q-learning, and model-assisted force feedback. A fixed protocol executes 34 policies on 120 fresh MuJoCo environments spanning geometry and physical-parameter shifts, plus 324 independently replayed action branc
Khai Loong Aw et al. · 借鉴儿童早期物理世界理解的发展特性，探讨零样本世界模型为何能数据高效地学习深度、运动与物体连贯性。属认知/世界模型交叉，未涉机器人操作实验。
Tatsuya Kamijo et al. · 提出TaMeSo-bot软体机器人，用掩码编码构建触觉记忆并结合柔性腕关节，在位置不确定下稳健完成键插入。接触丰富任务的触觉记忆设计可参考。
Zhiqi Li et al. · 把世界表示成可执行程序，通过递归场景程序从代码重建复杂3D世界。代码世界模型思路可用，但对象是3D场景重建而非机器人操作策略。
Jiadong Hong et al. · arXiv:2602.10420v4 Announce Type: replace Abstract: Recent work has popularized a practical recipe in diffusion and flow matching: predict the clean signal $x$, convert it to a velocity, and train through a velocity-space loss. The conversion contains a singular endpoint amplification and therefore appears prone to unstable optimization, yet recent systems obtain strong empirical results with this recipe. We investigate this tension through the integrability of the pre-optimizer stochastic-gradi