<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"
  xmlns:atom="http://www.w3.org/2005/Atom"
  xmlns:dc="http://purl.org/dc/elements/1.1/">
<channel>
  <title>Pulsar 照見 · VLA 每日信号</title>
  <link>https://sou350121.github.io/pulsar-web/daily</link>
  <atom:link href="https://sou350121.github.io/pulsar-web/rss/vla-daily.xml" rel="self" type="application/rss+xml" />
  <description>VLA 每日筛选的⚡🔧级论文 + SOTA 榜变动。仅保留 Pulsar 评级认为值得读的工作 · ❌和📖不进此 feed。</description>
  <language>zh-CN</language>
  <ttl>60</ttl>
  <lastBuildDate>Wed, 14 Oct 2026 09:04:22 GMT</lastBuildDate>
  <generator>Pulsar 照見 · https://github.com/sou350121/pulsar-web</generator>
  <copyright>Content under CC BY 4.0 — sou350121</copyright>
  <item>
    <title>🔧 VLMs are Native Action Models: Unleashing the Manipulation Capabilities of VLMs with Explicit Action-Space Grounding</title>
    <link>https://arxiv.org/abs/2610.12832v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.12832v1</guid>
    <pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：论证预训练 VLM 本身即具备操作所需的视觉/空间/推理能力，只需显式动作空间接地即可转为通用操作策略，挑战 VLA 必须额外训练动作头的假设。其动作接口设计可直接复用。

摘要：Pretrained vision-language models (VLMs) already exhibit strong visual, spatial and reasoning capabilities relevant to robotic manipulation. The key challenge in turning pretrained VLMs into generalist manipulation policies is to expose these capabilities through an appropriate action interface. Vision-language-action (VLA) models provide a unified trainable policy, but typically introduce learned action tokens, action heads, or action experts whose semantics must be learned from robot data pret

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Forward Dynamics Models</title>
    <link>https://arxiv.org/abs/2610.13117v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.13117v1</guid>
    <pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：对比 VLA（缺未来状态监督）与 WAM（生成模型计算昂贵）的局限，提出前向动力学模型作为统一框架，用未来世界状态监督策略学习。为 VLA/WAM 融合提供新范式，值得精读。

摘要：Vision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively. Yet VLAs learn without explicit supervision from future world states, while WAMs that condition actions on predicted futures incur extra inference cost and let visual prediction errors propagate into actions. We present Forward Dynamics Models (FDM), a family of robotic action models that

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Dual-Stream Memory for Vision-Language-Action Policies</title>
    <link>https://arxiv.org/abs/2610.13262v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.13262v1</guid>
    <pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：为 VLA 策略设计双流记忆：Gated DeltaNet 压缩执行状态，另一路恢复已不可观测的视觉证据，解决长时程记忆依赖操作中的状态跟踪与视觉回溯。可直接用于需记忆的操作策略改造。

摘要：Memory-dependent manipulation requires a policy to track execution state and recover visual evidence that is no longer observable. We propose dual-stream memory for vision-language-action policies, assigning these requirements to complementary representations. A task-agnostic Gated DeltaNet compresses history into one temporal token, supervised by progress prediction and action-conditioned feature reconstruction. A task-conditioned retriever retains frames above a fixed relevance threshold; line

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Noise Space Learning for VLA Adaptation</title>
    <link>https://arxiv.org/abs/2610.13943v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.13943v1</guid>
    <pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：提出噪声空间学习（NSL），在 flow-matching VLA 的噪声空间学习上下文条件化修正，实现少交互、低算力的任务适配。相比微调策略权重更省资源与更安全，可直接用于 π0 类 flow-matching VLA 的后训练。

摘要：Pretrained vision-language-action (VLA) models encode broad manipulation priors, yet adapting them to specific tasks requires balancing limited robot interaction, computational resources, and safety constraints. Noise space learning (NSL) adapts flow-matching-based VLAs by learning a context-conditioned distribution over noise while keeping the pretrained VLA fixed. Building on prior noise-steering methods, we formulate NSL as an interface for VLA adaptation that accommodates different learning 

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Splat2Act: From Captured Scenes to Personalized VLA Manipulation</title>
    <link>https://arxiv.org/abs/2610.13945v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.13945v1</guid>
    <pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：直击通用策略在用户自采场景部署失败（π0.5 在捕获场景拾取成功率 0%）的痛点，用场景捕获（3D 重建）把环境数据转化为个性化 VLA 操作策略。可复用于家庭/办公室真实部署场景的策略适配。

摘要：A user who brings a generalist manipulation policy into their home or office wants it to pick up their objects from their shelves, counters, and tables. Yet even a strong generalist like $π_{0.5}$ achieves 0% pickup success in our captured deployment scenes without adaptation, and collecting real demonstrations in every user's environment is prohibitively expensive. We present Splat2Act, an automated real-to-sim-to-real pipeline that personalizes a pretrained vision-language-action (VLA) policy 

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning</title>
    <link>https://arxiv.org/abs/2610.11945v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.11945v1</guid>
    <pubDate>Fri, 09 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：提出跨异构触觉传感器的人体触觉采集系统，把低成本人手手套数据映射到机器人触觉，缓解触觉演示稀缺；直接服务触觉 VLA 数据扩展。

摘要：Collecting tactile demonstrations on robots is costly and slow, motivating the use of lower-cost human tactile gloves for scalable data collection. However, human capacitive/piezoresistive gloves and robotic tactile sensors differ fundamentally in transduction principle, sensor layout, spatial resolution, and dynamic response, making alignment of raw sensor channels ill-posed. To address this problem, we present TACROSS, a scalable system for learning from human touch and transferring it to robo

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models</title>
    <link>https://arxiv.org/abs/2610.12007v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.12007v1</guid>
    <pubDate>Fri, 09 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：针对 flow-based VLA 提出滚动去噪与双重解耦，缓解长动作块平滑与频繁重规划反应性之间的矛盾；直接用于 VLA 闭环部署。

摘要：Flow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning improves reactivity at the cost of action discontinuities. We introduce REACT, a rolling-denoising framework that makes flow-based VLAs more reactive while preserving long-horizon context. Instead of regenerating entire action chunks from scratch, REACT maintai

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 PLaW-VLA: Predictive Latent World Modeling for Vision-Language-Action Policies</title>
    <link>https://arxiv.org/abs/2610.12285v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.12285v1</guid>
    <pubDate>Fri, 09 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：为 VLA 引入预测性潜在世界建模，只建模任务相关的未来表征并据此条件化动作生成，改善长时序控制；可直接接入现有 VLA 策略。

摘要：Learning to predict how the world evolves can provide vision-language-action (VLA) policies with predictive context for long-horizon control, but its effectiveness depends on what future representation is modeled and how it conditions action generation. We introduce PLaW-VLA, which models task-relevant future states in a pretrained prediction-oriented representation space, reducing the need to predict control-irrelevant visual details. Built on a Mixture-of-Transformers architecture, PLaW-VLA co

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 ARC: A Reasoning Recipe for Robot Foundation Models</title>
    <link>https://arxiv.org/abs/2610.12386v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.12386v1</guid>
    <pubDate>Fri, 09 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[NVIDIA|Ramos]</category>
    <description><![CDATA[理由：提出面向机器人基础模型的推理后训练配方，无需更大模型或更多数据即可显著提升零样本任务表现；可直接复用到现有 VLA 后训练流程。

摘要：The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale. We show that there exists an effective and efficient complementary approach: the right reasoning recipe can substantially improve the zero-shot task performance of existing state-of-the-art RFMs. We refer to this recipe as ARC. It consists of three key ingredients: a reasoning trace, a scalable automatic labeling pipeline, and a strategy for adapti

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation</title>
    <link>https://arxiv.org/abs/2610.12451v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.12451v1</guid>
    <pubDate>Fri, 09 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：提出相机配置可变的 VLA 策略，在部署时相机数量/位姿变化下依然鲁棒，直击固定相机训练导致的部署脆弱性；可直接用于实际 VLA 部署。

摘要：Vision-Language-Action (VLA) models have emerged as powerful foundations for robotic manipulation, but their reliance on fixed camera configurations during training makes them brittle to changes in camera count or pose during deployment. To overcome these limitations, we propose VersaCamVLA, a camera-configurable framework that decouples camera-set representation from action learning. VersaCamVLA learns a unified scene-token interface that maps an arbitrary, variable set of posed RGB views into 

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning</title>
    <link>https://arxiv.org/abs/2610.10288v2</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.10288v2</guid>
    <pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：发布 500 小时第一人称人类视觉-触觉交互数据，填补纯视频缺失接触与压力监督的空白。为视觉-触觉 VLA 预训练提供大规模物理监督数据，可直接下载使用。

摘要：Large-scale egocentric human interaction data is becoming an important source of physical supervision for embodied learning, yet video alone leaves the contact and pressure that characterize physical interaction unrecorded. Recent visual-tactile datasets provide this missing supervision, but their synchronized tactile data remain far smaller in volume than human video. Moreover, the largest resources often merge recordings from different sensors or annotation procedures, which makes the effect o

来源：cs.CV]]></description>
  </item>
  <item>
    <title>🔧 Juno: Taming Predictive Latents for Vision-Language-Action Models</title>
    <link>https://arxiv.org/abs/2610.09940v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.09940v1</guid>
    <pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：将 JEPA 式联合嵌入预测架构的预测隐变量引入 VLA，统一预训练、策略学习与部署三阶段的隐表征使用。为 VLA 引入预测式表征提供可复用的训练范式。

摘要：Joint-embedding predictive architectures (JEPAs) predict masked or future observations in representation space, offering a natural source of predictive latents for vision-language-action (VLA) models. Yet making these latents useful across pretraining, policy learning, and deployment requires addressing three failures: mismatch with embodiment-specific control, interference with action learning, and teacher miscalibration under distribution shifts. We introduce Juno, a unified framework built ar

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Many Ways to Succeed: Diversity-Driven RL Fine-Tuning for VLA Generalization</title>
    <link>https://arxiv.org/abs/2610.09943v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.09943v1</guid>
    <pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：分析 RL 微调会全局收缩探索仅在成功区多样化的现象，提出多样性驱动的 RL 微调提升 VLA 在微调分布外的泛化。做 VLA RL 精调可直接借鉴其探索重塑策略。

摘要：Reinforcement learning (RL) fine-tuning improves vision-language-action (VLA) policies through closed-loop experience, yet generalization beyond the fine-tuning distribution remains limited. Our analysis reveals a selective reshaping of exploration: RL contracts behavior globally, yet diversifies successful trajectories, elicits success with fewer rollouts, and covers more of the latent task-valid solution space than supervised fine-tuning. Broader successful-mode coverage may provide alternativ

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 OpenViTac: Learning and Benchmarking Visuo-Tactile Policies in a Unified Sim-and-Real Framework</title>
    <link>https://arxiv.org/abs/2610.10384v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.10384v1</guid>
    <pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：提出统一的仿真-真实框架，用于学习并基准评测视触觉-语言-动作（VTLA）策略，填补触觉 VLA 缺少统一评测基准的空白。做触觉操作可直接采用其评测与训练流程。

摘要：Tactile feedback provides embodied agents with physical information beyond visual observations, enabling more reliable interaction with the real world. However, despite the rapid progress of vision-tactile-language-action (VTLA) policies, there remains a lack of unified benchmarks for evaluating tactile-enabled robot manipulation across simulation and the real world. To address this gap, we introduce OpenViTac, a visuo-tactile manipulation benchmark for evaluating robot policies across simulatio

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Rephrase Before You Act: Characterizing and Mitigating Language Sensitivity in Vision-Language-Action Models</title>
    <link>https://arxiv.org/abs/2610.10526v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.10526v1</guid>
    <pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[LIBERO Team]</category>
    <description><![CDATA[理由：揭示 VLA 对指令措辞极度敏感（如 π0.5 在 LIBERO 上换词即成功率骤变），且未继承底层 VLM 的语言鲁棒性，并提出改写指令的缓解方法。做语言泛化/指令鲁棒性可直接复用。

摘要：Vision-language-action models (VLAs) are strikingly sensitive to instruction phrasing and do not inherit the language robustness of the vision-language models they are built on. A one-word edit can move success by tens of points: $π_{0.5}$ turns on a LIBERO stove 100% of the time for "switch on the stove" and 2% for "switch on the hot plate", and a $π_0$ checkpoint finetuned with rephrase augmentation still shows swings of up to 61 points. We characterize this sensitivity with statistically test

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Commit While Futures Agree: Consequence-Aware Adaptive Action Chunking for Robot Manipulation</title>
    <link>https://arxiv.org/abs/2610.07949v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.07949v1</guid>
    <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：提出“未来预测一致才继续执行”的自适应 action chunk 提交长度，取代固定执行前缀，可直接插入任意 action-chunking VLA 策略。

摘要：Action-chunking policies predict multi-step control sequences, but a fundamental question remains: how much of a predicted action chunk should be committed before replanning? Existing systems typically execute a fixed-length prefix, implicitly assuming that the same execution horizon remains trustworthy across states. Some adaptive methods estimate this horizon from the similarity or stability of predicted actions. However, different actions may lead to the same successful outcome, whereas simil

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 EgoLAP: Learning from Egocentric Human Data through Language-Action Reasoning</title>
    <link>https://arxiv.org/abs/2610.08726v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.08726v1</guid>
    <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[Stanford|Sadigh]</category>
    <description><![CDATA[理由：用语言-动作推理桥接人类第一视角数据与机器人控制：低层动作虽本体相关，其运动意图可跨本体监督。为人类数据规模化训练 VLA 提供可复用路径。

摘要：Egocentric human data offer a path to scaling robot learning beyond costly robot demonstrations, yet the embodiment gap makes raw human trajectories a poor supervisory target for control. Our key insight is that, although low-level actions are embodiment-specific, their underlying motion intent can capture task-relevant structure that transfers across humans and robots. We introduce EgoLAP, a VLA pre-training framework that jointly learns from human and robot trajectories through a shared langua

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 PEARS: Physical-Prior-Guided Efficient Adaptation via Failure Reasoning and Diffusion Steering for Tactile Manipulation</title>
    <link>https://arxiv.org/abs/2610.08784v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.08784v1</guid>
    <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>read-only</category>
    <description><![CDATA[理由：面向触觉操作的 OOD 后训练框架：用物理先验与失败推理引导 diffusion steering，避免大规模真机 RL 采样。可复用其失败驱动的在线纠偏流程到触觉 VLA。

摘要：Pretrained robotic policies can suffer substantial performance degradation under out-of-distribution (OOD) conditions encountered during deployment, motivating post-training through real-world interaction. However, reinforcement-learning (RL)-based post-training typically requires substantial environment interactions, a burden that is especially significant in manipulation, where each trial can be slow, costly, or destructive. Therefore, we present PEARS, a physics-prior-guided hybrid RL framewo

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Beyond Reconstruction: What Matters in Action Tokenization for Robot Policies?</title>
    <link>https://arxiv.org/abs/2610.09170v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.09170v1</guid>
    <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[TTIC|Walter]</category>
    <description><![CDATA[理由：系统对比重建式与行为导向的 action tokenizer，指出重建目标与闭环控制不匹配，为自回归 VLA 的动作分词器选型提供可复用准则与实验结论。

摘要：Autoregressive action-token policies such as vision-language-action models require action tokenizers to translate discrete token sequences into precise control actions in continuous space. Many action tokenizers learn the mapping between tokens and actions via a reconstruction objective. However, as we show through extensive analysis, sufficiently accurate action reconstruction is only one part of what makes a downstream robot policy successful. It is also critical that the policy is able to pre

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Co-Evolving Robot Orchestrators and Policies through Deployment</title>
    <link>https://arxiv.org/abs/2610.09228v1</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.09228v1</guid>
    <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[Stanford|Wu]</category>
    <description><![CDATA[理由：让 VLM orchestrator 与 VLA 策略在真实部署中协同演化：用失败轨迹自动生成新任务与纠正数据，闭环迭代提升开放环境泛化。可复用其 orchestrator-policy 协同训练闭环设计。

摘要：Vision-language-action (VLA) policies trained on large datasets are capable within their training domains, yet they still fail to generalize to the variety of situations a robot meets in real-world deployment. Agentic robot systems complement the policy with a vision-language model (VLM) orchestrator that learns when to call the policy, how to instruct it, and when to use scripted skills instead. However, because the harness is built around a frozen policy that has limited language steerability,

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Equivariant Visual-Tactile Diffusion Policy for Contact-Rich Manipulation</title>
    <link>https://arxiv.org/abs/2610.03333</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.03333</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：提出工作空间级等变的视觉-触觉扩散策略 VISTA，利用对称性等变降低专家数据需求，提升接触丰富操作的样本效率，可直接复用于触觉 VLA 的数据高效训练。

摘要：arXiv:2610.03333v1 Announce Type: new Abstract: Imitation learning for contact-rich manipulation requires high-quality expert data that is expensive to obtain. This makes learning a sample-efficient policy a key issue. To address this, we propose VISTA, a workspace-level equivariant visuotactile diffusion policy for data-efficient contact-rich imitation learning. VISTA projects visual and tactile observations into spherical tokens, injects tactile contact cues into visual spherical directions th

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 FastOPD: On-Policy Distillation for Lightweight VLA Deployment</title>
    <link>https://arxiv.org/abs/2610.02832</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.02832</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：针对 VLA 基础模型部署算力瓶颈，提出 on-policy 蒸馏将大 VLA 压缩为轻量策略，兼顾操作性能与推理效率，为 VLA 边缘/实时部署提供可落地的压缩方案。

摘要：arXiv:2610.02832v1 Announce Type: new Abstract: Vision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging. Existing approaches typically mitigate this issue by designing smaller architectures or reducing the iterative denoising steps in flow-based policies. In this work, we propose FastOPD, a foundation-to-lightweight VLA f

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 SimpleTouch: Can Vision-Language-Action Models Master Contact-Rich Manipulation Without Tactile Policy Pretraining?</title>
    <link>https://arxiv.org/abs/2610.02784</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.02784</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：系统探究 VLA 模型能否在没有触觉策略预训练的情况下掌握接触丰富操作，给出简化的触觉注入方案，对触觉 VLA 是否需专门预训练这一关键问题给出直接答案，可指导触觉微调流程。

摘要：arXiv:2610.02784v1 Announce Type: new Abstract: Tactile sensing provides essential contact information for robotic manipulation, yet incorporating it into pretrained vision-language-action (VLA) models remains challenging. A common concern is that simply introducing touch during task-specific fine-tuning may fail to bridge the cross-modal gap, yielding limited gains or even reduced success. Consequently, existing methods often rely on large-scale tactile policy pretraining or separate visuotacti

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 Rethinking World-Action Model for Compositional and In-Context Robotic Manipulation</title>
    <link>https://arxiv.org/abs/2610.02368</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.02368</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：针对长时程组合操作改造世界-动作模型，解决短时程视觉未来预测与多子任务协调的割裂，支持组合任务与上下文内学习。可复用于 WAM 的长时程训练与任务分解。

摘要：arXiv:2610.02368v1 Announce Type: new Abstract: Long-horizon compositional manipulation has become increasingly important for real-world robot deployment, where a single task involves multiple coordinated subtasks. Existing world-action models (WAMs) jointly predict short-horizon visual futures and actions, but typically lack explicit subtask-level reasoning. We propose Visual Goal-conditioned Action Reasoning (ViGAR), a hierarchical framework that factorizes manipulation into a visual subgoal p

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models</title>
    <link>https://arxiv.org/abs/2610.02323</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.02323</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <description><![CDATA[理由：针对流匹配 VLA 从任务无关高斯源采样、忽略执行历史的问题，用世界模型校准的 proposal-to-action 流重构源分布以提升动作块生成。可直接复用于 flow-based VLA 的源建模。

摘要：arXiv:2610.02323v1 Announce Type: new Abstract: Flow-based Vision-Language-Action (VLA) policies generate action chunks by transporting samples from a task-agnostic isotropic Gaussian source. As this source is conditioned on neither recent execution nor predicted future evolution, (i) it discards the local continuity established by recently executed motion. (ii) Even when predictive world representations are introduced, they often only condition the transport dynamics rather than determine where

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 eRLT: Efficient VLA Reinforcement Learning via Action-Relevant Token Routing</title>
    <link>https://arxiv.org/abs/2610.00913</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.00913</guid>
    <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[Southern University of Science and Technology, Shenzhen, China]</category>
    <description><![CDATA[理由：通过动作相关 token 路由实现高效 VLA 强化学习，降低在线 RL 的计算开销；为冻结 VLA 的下游适配提供更省算力的 RL 微调路径。

摘要：arXiv:2610.00913v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models provide strong behavioral priors for robotic manipulation, yet efficiently adapting them to downstream tasks remains challenging. Recent work addresses this challenge by adapting frozen VLAs through online reinforcement learning (RL), whose sample efficiency depends on the quality of the state representation used by the actor and critic. Existing methods construct such representations either with VLA-independent

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models</title>
    <link>https://arxiv.org/abs/2610.00899</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.00899</guid>
    <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[AIST]</category>
    <description><![CDATA[理由：面向自回归 VLA 提出随机动作 tokenization，在 FAST 之上用随机采样编码连续动作以提升离散动作表征表达力；可直接替换现有 VLA 的动作 tokenizer。

摘要：arXiv:2610.00899v1 Announce Type: new Abstract: Autoregressive Vision-Language-Action models often represent continuous robot actions as discrete token sequences, enabling action prediction with standard next-token objectives. FAST has substantially improved this representation by compactly encoding action containing diverse temporal frequencies into relatively few tokens. However, while such compression reduces the number of action tokens required for autoregressive prediction, it does not nece

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 ECoMEM: Explicit Concept Memory for Memory-Dependent Robot Control</title>
    <link>https://arxiv.org/abs/2610.00801</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.00801</guid>
    <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[Stanford]</category>
    <description><![CDATA[理由：为记忆依赖型控制引入显式概念记忆，让 VLA 记住已消失的物体、前人演示与任务进度；外部记忆模块可直接接入现有 VLA 策略。

摘要：arXiv:2610.00801v1 Announce Type: new Abstract: A robot may lose sight of an object it must later retrieve, need to recall what a person demonstrated earlier, or track which steps of a task it has already completed. Current vision-language-action (VLA) policies often fail once the information needed for action disappears from the current observation, making memory critical for long-horizon robot behavior. Existing approaches typically provide longer histories or learn implicit memory from observ

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 TacDyn-WAM: Learning Implicit Tactile Dynamics in a Heterogeneous Visuo-Tactile World Action Model</title>
    <link>https://arxiv.org/abs/2610.00638</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.00638</guid>
    <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[Institute for AI Industry Research (AIR), Tsinghua University]</category>
    <description><![CDATA[理由：在异质视触觉 world-action model 中隐式学习触觉动力学，跳过视频生成式迭代去噪，直接以动作条件预测未来触觉表征；触觉 VLA 可复用其视触融合与动力学建模路径。

摘要：arXiv:2610.00638v1 Announce Type: new Abstract: World action models improve robotic manipulation by conditioning actions on predicted futures, yet existing tactile variants largely inherit video-generation pipelines that reconstruct future tactile observations through iterative denoising. Such prediction can become unreliable under deployment drift: small changes in contact position or force may substantially alter tactile pixels even when the underlying contact evolution remains predictable. We

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🔧 DriftOPD: Sequence-Level Reverse-KL Distillation for One-Step VLA Policies</title>
    <link>https://arxiv.org/abs/2610.00317</link>
    <guid isPermaLink="false">hotspot:https://arxiv.org/abs/2610.00317</guid>
    <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>

    <category>VLA 每日热点</category>
    <category>actionable</category>
    <category>[KAIST]</category>
    <description><![CDATA[理由：把 VLA 动作专家蒸馏为单步策略，用序列级 reverse-KL 替代逐 chunk 似然训练，缓解 receding-horizon 推理延迟；可直接用于 π0/OpenVLA 类动作专家的加速部署。

摘要：arXiv:2610.00317v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models increasingly rely on action experts that generate short action chunks under receding-horizon control. While chunk-level training is convenient across robot embodiments, it optimizes local action likelihood without explicitly accounting for long-horizon task success. Sequence-level reinforcement learning can address this limitation, but typically requires policy rollouts and closed-loop interaction, which are cost

来源：cs.RO]]></description>
  </item>
  <item>
    <title>🏆 SOTA: Evo-World on MetaWorld — average 91.1</title>
    <link>https://sota.evomind-tech.com/benchmarks/metaworld/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:MetaWorld:Evo-World:2026-10-12</guid>
    <pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>MetaWorld</category>
    <description><![CDATA[Benchmark：MetaWorld · split=standard-opensource
指标：average = 91.1
模型：Evo-World · 2610.12936]]></description>
  </item>
  <item>
    <title>🏆 SOTA: Light-O1 on RoboCasa-GR1-Tabletop — avg_success_rate 79.3</title>
    <link>https://sota.evomind-tech.com/benchmarks/robocasa_gr1_tabletop/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:RoboCasa-GR1-Tabletop:Light-O1:2026-10-04</guid>
    <pubDate>Sun, 04 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>RoboCasa-GR1-Tabletop</category>
    <description><![CDATA[Benchmark：RoboCasa-GR1-Tabletop · split=standard-opensource
指标：avg_success_rate = 79.3
模型：Light-O1 · light-o1]]></description>
  </item>
  <item>
    <title>🏆 SOTA: AVA-VLA on CALVIN — avg_len 4.65</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:AVA-VLA:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=ABC-D
指标：avg_len = 4.65
模型：AVA-VLA · 2511.18960]]></description>
  </item>
  <item>
    <title>🏆 SOTA: MMaDA-VLA on CALVIN — avg_len 4.78</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:MMaDA-VLA:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=ABC-D
指标：avg_len = 4.78
模型：MMaDA-VLA · 2603.25406]]></description>
  </item>
  <item>
    <title>🏆 SOTA: NS-VLA on CALVIN — avg_len 4.56</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:NS-VLA:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=ABC-D
指标：avg_len = 4.56
模型：NS-VLA · 2603.09542]]></description>
  </item>
  <item>
    <title>🏆 SOTA: EDAR on CALVIN — avg_len 4.75</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:EDAR:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=ABCD-D
指标：avg_len = 4.75
模型：EDAR · 2607.11427]]></description>
  </item>
  <item>
    <title>🏆 SOTA: Xiaomi-Robotics-0 on CALVIN — avg_len 4.8</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:Xiaomi-Robotics-0:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=ABCD-D
指标：avg_len = 4.8
模型：Xiaomi-Robotics-0 · 2602.12684]]></description>
  </item>
  <item>
    <title>🏆 SOTA: HiMoE-VLA on CALVIN — avg_len 4.49</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:HiMoE-VLA:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=D-D
指标：avg_len = 4.49
模型：HiMoE-VLA · 2512.05693]]></description>
  </item>
  <item>
    <title>🏆 SOTA: MCIL on CALVIN — avg_len 1.82</title>
    <link>https://sota.evomind-tech.com/benchmarks/calvin/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:CALVIN:MCIL:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>CALVIN</category>
    <description><![CDATA[Benchmark：CALVIN · split=D-D
指标：avg_len = 1.82
模型：MCIL · 2005.07648]]></description>
  </item>
  <item>
    <title>🏆 SOTA: PLD on LIBERO — average 99.17</title>
    <link>https://sota.evomind-tech.com/benchmarks/libero/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:LIBERO:PLD:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>LIBERO</category>
    <description><![CDATA[Benchmark：LIBERO · split=non-standard
指标：average = 99.17
模型：PLD · 2511.00091]]></description>
  </item>
  <item>
    <title>🏆 SOTA: FutureDuet on LIBERO — average 99.2</title>
    <link>https://sota.evomind-tech.com/benchmarks/libero/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:LIBERO:FutureDuet:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>LIBERO</category>
    <description><![CDATA[Benchmark：LIBERO · split=standard-closed
指标：average = 99.2
模型：FutureDuet · 2609.34362]]></description>
  </item>
  <item>
    <title>🏆 SOTA: LaST-R1 on LIBERO — average 99.8</title>
    <link>https://sota.evomind-tech.com/benchmarks/libero/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:LIBERO:LaST-R1:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>LIBERO</category>
    <description><![CDATA[Benchmark：LIBERO · split=standard-opensource
指标：average = 99.8
模型：LaST-R1 · 2604.28192]]></description>
  </item>
  <item>
    <title>🏆 SOTA: CorridorVLA on LIBERO Plus — total 83.21</title>
    <link>https://sota.evomind-tech.com/benchmarks/liberoplus/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:LIBERO Plus:CorridorVLA:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>LIBERO Plus</category>
    <description><![CDATA[Benchmark：LIBERO Plus · split=non-standard
指标：total = 83.21
模型：CorridorVLA · 2604.21241]]></description>
  </item>
  <item>
    <title>🏆 SOTA: Qwen-RobotManip on LIBERO Plus — total 91.4</title>
    <link>https://sota.evomind-tech.com/benchmarks/liberoplus/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:LIBERO Plus:Qwen-RobotManip:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>LIBERO Plus</category>
    <description><![CDATA[Benchmark：LIBERO Plus · split=standard-closed
指标：total = 91.4
模型：Qwen-RobotManip · 2606.17846]]></description>
  </item>
  <item>
    <title>🏆 SOTA: FabriMAE on LIBERO Plus — total 86.8</title>
    <link>https://sota.evomind-tech.com/benchmarks/liberoplus/?utm_source=rss&amp;utm_medium=feed&amp;utm_campaign=vla-daily-sota</link>
    <guid isPermaLink="false">sota:LIBERO Plus:FabriMAE:2026-10-02</guid>
    <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>

    <category>SOTA 榜</category>
    <category>LIBERO Plus</category>
    <description><![CDATA[Benchmark：LIBERO Plus · split=standard-opensource
指标：total = 86.8
模型：FabriMAE · 2608.16697]]></description>
  </item>
</channel>
</rss>