Embodied AI Evidence Hub具身智能证据知识库

数据工程与质量 · Data Engineering and Quality

具身智能数据质量的主要矛盾

Core Contradictions in Embodied AI Data Quality

更新于

30 秒结论

先读这个判断

版本说明：本轮以 15 篇可获取完整正文的论文为论证主干，逐篇核对问题、方法、结果与限制；未能取得可读全文的论文不再承担正文结论。

精读论文
15
正式证据事件
15
知识单元
EA-DATA, EA-SENSOR, EA-HARDWARE, EA-XEMBODIMENT, EA-MODEL, EA-EVAL
审计状态
当前结算 run 已通过仓库校验

v0 · 完整正文

研究综述

研究边界

版本说明：本轮以 15 篇可获取完整正文的论文为论证主干，逐篇核对问题、方法、结果与限制；未能取得可读全文的论文不再承担正文结论。

本文覆盖 2026 年 1 月 14 日至 7 月 14 日公开论文，从 822 条去重候选中核验 124 篇可用全文，并以 33 篇直接相关论文支撑示教筛选、四维几何、触觉世界模型、跨本体对齐和闭环评测等判断。它讨论的是数据何时能改善机器人策略，而不是建立一个脱离任务的通用质量排行榜。覆盖足以识别共同矛盾，但不同论文的任务、机器人和评测口径并不统一。

中心判断

具身智能数据质量的主要矛盾，已经从“数据少与模型大”转变为“规模扩张与因果可用性”之间的冲突。机器人真正需要的不是更多观测，而是能把任务意图、三维状态、动作语义、接触后果和失败恢复连成闭环的数据。数据规模越大，若模态可靠性、本体差异和动作结果没有被保存，错误监督也会按同样速度放大。

矛盾一：数量增加，行为结构反而被稀释

SIEVE 表明，示教中的冗余、噪声和覆盖不均并非中性成本；保留可复用行为结构后，较少数据可以超过全量训练。这个结果否定了“只要继续堆轨迹，模型会自动平均掉坏样本”的直觉。重复动作会放大常见模式，稀有恢复和边界状态却可能在总量中被淹没。

但筛选也不能退化成只保留最高分轨迹。多任务训练需要守住任务、本体、场景和失败类型的覆盖，否则局部效用最高的数据会制造新的分布偏差。质量治理因此是约束优化：一边减少有害冗余，一边保护对泛化有价值的差异。

矛盾二：视觉覆盖很广，物理状态仍然不可见

Lift3D-VLA 和 GEM-4D 都指向二维视觉的边界：看清物体类别不等于掌握度量几何，生成逼真画面也不等于保持跨帧三维对应。对于遮挡、形变和接触，数据必须显式保存空间结构与时间连续性，模型才能区分“物体看起来没变”与“接触状态已经失稳”。

Deform360 进一步说明，软物体数据需要多视角全局运动、局部触觉形变和可评测三维轨迹共同约束。TACO 则把纯视觉纠错的失败落到接触扰动上：图像可以合理，触觉却已不一致。多模态不是装饰，而是补足任务状态的可观测性。

矛盾三：异构数据扩大覆盖，也破坏监督等价性

人类第一视角视频、手持接口、真实机器人示教和生成式轨迹都能扩展数据来源，却不能被当成同一种真值。τ0-WM 将它们分配给不同监督角色：人类视频提供视觉动力学，机器人示教锚定可执行动作，失败轨迹训练纠错。若缺失模态被强行补成标签，数据混合会把不确定性伪装成确定监督。

跨本体问题更明显。相同动作命令在不同控制器、坐标系和硬件上可能产生不同运动；只统一文件格式，不能统一动作含义。数据资产必须保留本体、控制频率、坐标约定和接触条件，训练时再通过适配器或共享中间表示对齐。

矛盾四：离线指标易算，闭环价值难证

GigaWorld-1 提醒，世界模型的短期视觉质量不等于策略评估可靠性；真正关键的是动作条件下的长程一致性。示教数据同样如此：轨迹平滑、画面清晰和标签完整都只是代理指标，最终要回答它是否提高真实成功率、恢复能力和分布外稳定性。

这要求质量体系形成反馈回路。采集阶段发现的停顿和关节极限应返回给采集员；训练阶段识别出的高影响片段应调整采样；部署失败又应反向生成困难样本。没有闭环，质量评分会停在静态报表，无法适应目标策略变化。

条件、分歧与风险

现有证据不支持一套跨任务通用阈值。伪标注在学习相对几何时可能足够，在精密装配中却可能不可接受；触觉对接触任务关键，在开放空间移动中收益有限；数字遥操作能降低硬件成本，却仍受复杂物理和本体微调约束。因此，每个质量指标都必须声明任务、模态、时间尺度和可接受风险。

数据治理也不能把所有失败数据删掉。失败附近常含最有价值的恢复信息，次优长轨迹中可能存在可复用片段。更稳妥的做法是记录片段级进度和失败类型，再决定过滤、降权、分阶段训练或专门用于纠错。

可操作框架与下一步

建议把质量账本拆成五列：采集接口是否引入系统偏差，状态是否可观测且同步，动作语义是否与本体一致，样本是否补充目标覆盖，训练后是否带来闭环收益。每列都需要可复现的检测与反例，而不是汇总成一个不可解释的总分。

下一步研究应建设跨任务的干预式评测：固定模型与训练预算，只改变数据筛选、模态组合或本体对齐策略，再比较真实动作结果。只有这样，才能把“数据看起来更好”推进为“数据导致策略更可靠”。

结论

具身数据质量不是清洗环节，而是从采集硬件到部署反馈的系统属性。规模仍然重要，但只有当数据保持可观测状态、可解释动作和可验证后果时，规模才会转化为能力。未来的数据竞争不会只比小时数，而会比谁更早发现无效监督、谁能保留困难状态、谁能用闭环结果持续重写采集与训练规则。

组织与治理含义

上述矛盾还改变了数据团队的组织方式。采集、算法和部署不能各自优化局部指标：采集员需要知道哪些失败状态最稀缺，算法团队需要看到硬件与操作者带来的系统偏差，部署团队则要把接管、碰撞和恢复日志返还给数据管线。只有共享同一质量账本，问题才不会在部门之间被反复改名。

质量版本也应与模型版本绑定。目标任务、动作空间或本体发生变化后，旧排序不再天然有效；同一条轨迹可能从高价值变成不兼容。每次训练都应记录使用的数据快照、筛选规则、权重和排除原因，使策略退化能够追到具体数据决策。

成本评估不能只算采集小时。结构筛选需要计算与验证，多模态需要标定和维护，真实失败数据还涉及安全成本。更合理的指标是单位闭环收益所需的总成本，包括采集、清洗、标注、验证与机器人占用时间。这样才能比较“多采”“精筛”和“补模态”究竟哪条路线更经济。

References

Accepted evidence

去重论文引用

  1. Deform360: A Massive Multi-view Visuotactile Dataset for Deformable World ModelsHongyu Li, Wanjia Fu, Xiaoyan Cong, Zekun Li, Binghao Huang, Hanxiao Jiang, Xintong He, Yiqing Liang et al. · 2026
  2. From Interaction to Demonstration Quality in Virtual Reality: Effects of Interaction Modality and Visual Representation on Everyday TasksRobin Beierling, Manuel Scheibl, Jonas Dech, Abhijit Vyas, Anna-Lisa Vollmer · 2026
  3. GEM-4D: Geometry-Enhanced Video World Models for Robot ManipulationKaichen Zhou, Yuzhen Chen, Fangneng Zhan, Hang Hua, Grace Chen, Xinhai Chang, Ao Qu, Yilun Du et al. · 2026
  4. GigaWorld-1: A Roadmap to Build World Models for Robot Policy EvaluationGigaWorld Team, Angyuan Ma, Boyuan Wang, Bohan Li, Chaojun Ni, Guo Li, Guan Huang, Guosheng Zhao et al. · 2026
  5. Imitating What Works: Simulation-Filtered Modular Policy Learning from Human VideosAlbert J. Zhai, Kuo-Hao Zeng, Jiasen Lu, Ali Farhadi, Shenlong Wang, Wei-Chiu Ma · 2026
  6. Influence of Gripper Design on Human Demonstration Quality for Robot LearningGina L. Georgadarellis, Natalija Beslic, Seonhun Lee, Frank C. Sup, Meghan E. Huber · 2026
  7. Lift3D-VLA: Lifting VLA Models to 3D Geometry and Dynamics-Aware ManipulationJiaming Liu, Qingpo Wuwu, Nuowei Han, Hao Chen, Zhuoyang Liu, Fan Fei, Yueru Jia, Chenyang Gu et al. · 2026
  8. RoboDream: Compositional World Models for Scalable Robot Data SynthesisJunjie Ye, Rong Xue, Basile Van Hoorick, Runhao Li, Harshitha Rajaprakash, Pavel Tokmakov, Muhammad Zubair Irshad, Vitor Guizilini et al. · 2026
  9. RynnWorld-Teleop: An Action-Conditioned World Model for Digital TeleoperationHaoyu Zhao, Xingyue Zhao, Hangyu Li, Biao Gong, Kehan Li, Siteng Huang, Xin Li, Deli Zhao et al. · 2026
  10. SIEVE: Structure-Aware Data Selection for Imitation Learning with VLA ModelsChangti Wu, Bin Yu, Zhaolong Shen, Shijie Lian, Xiaopeng Lin, Cong Huang, Zhirui Zhang, Lei Zhang et al. · 2026
  11. TACO: TActile World Model as a Self-COrrector forScalable VLA Post-TrainingShengbang Liu, Yueru Jia, Yuyang Yan, Jiaming Liu, Xinran Zhang, Qiuxuan Feng, Yandong Guo, Shiji Zhou et al. · 2026
  12. WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data CurationJustin Yu, Andrew Goldberg, Kavish Kondap, Karim El-Refai, Ethan Ransing, Qianzhong Chen, Mac Schwager, Fred Shentu et al. · 2026
  13. World Model Self-Distillation: Training World Models to Solve General TasksSebastian Stapf, Pablo Acuaviva Huertos, Aram Davtyan, Paolo Favaro · 2026
  14. World Pilot: Steering Vision-Language-Action Models with World-Action PriorsZefu Lin, Rongxu Cui, Junjia Xu, Xiaojuan Jin, Wenling Li, Lue Fan, Zhaoxiang Zhang · 2026
  15. τ0-WM: A Unified Video-Action World Model for Robotic ManipulationPengfei Zhou, Shengcong Chen, Di Chen, Jiaxu Wang, Rongjun Jin, Bingwen Zhu, Yike Pan, Songen Gu et al. · 2026
完整证据附录（15 条正式证据事件）
  • Time range: 2026-01-14..2026-07-14
  • Events: 15
  • 每个事件一节,标题即锚点;trace-map 中的 event 链接跳转到这里。

EA-DQ-CONTRA-READ-0008

  • Claim: 示教数据质量会被采集硬件本身塑形；UMI 类手持 gripper 的力分布、重量和人体工学会影响任务表现、操作者负担和后续可学习策略。
  • Stance: support | Confidence: direct
  • Paper: 2603.17189 Influence of Gripper Design on Human Demonstration Quality for Robot Learning
  • Locator: V DISCUSSION
  • Evidence: 论文指出 UMI 示教虽快于遥操作但仍比手工慢、工具重量会造成疲劳并影响 demonstration；实验中改变 UMI gripper fingers 的力分布显著影响打开绷带包装表现，concentrated load grippers 优于 distributed load grippers，作者将其连接到 demonstration quality 和 learned robot control policies。
  • Quote: “Overall, the usability study demonstrated that altering the force distribution of UMI gripper fingers significantly affected participants’ ability to open bandage packages, with concentrated load grippers outperforming distributed load grippers. These findings highlight that subtle hardware changes can substantially improve demonstration quality and, in turn, the robot control policies learned from them.”
  • Authors: gina-l-georgadarellis; natalija-beslic; seonhun-lee; et al.

EA-DQ-CONTRA-READ-0003

  • Claim: World-model training data needs geometry-consistency supervision, because photorealistic video without stable 4D correspondences can fail to yield executable robot actions.
  • Stance: support | Confidence: direct
  • Paper: 2605.22882 GEM-4D: Geometry-Enhanced Video World Models for Robot Manipulation
  • Locator: Abstract (full-text section)
  • Evidence: GEM-4D injects dense 4D correspondence supervision from a geometry foundation model into a video generative backbone during training, arguing that correspondence consistency makes future rollouts more reliable for action extraction.
  • Quote: “Abstract Video world models can generate realistic futures from a single instruction, but they often fail to track the same physical points consistently across time. As a result, the generated videos appear plausible, yet lack the physical grounding required for reliable action execution, such as robot manipulation. We present GEM-4D , a geometry-grounded video world model that resolves this limitation by injecting dense 4D correspondence supervision distilled from a pretrained geometry foundati”
  • Authors: kaichen-zhou; yuzhen-chen; fangneng-zhan; et al.

EA-DQ-CONTRA-READ-0007

  • Claim: τ0-WM 使用真实机器人遥操作、UMI 式交互、人类第一视角视频与 rollout/失败轨迹的异构语料，并用按模态的监督掩码训练统一视频—动作世界模型。
  • Stance: support | Confidence: direct
  • Paper: 2606.01027 τ0-WM: A Unified Video-Action World Model for Robotic Manipulation
  • Locator: Abstract (full-text section)
  • Evidence: 摘要直接报告了异构数据组成与 modality-specific supervision masks。
  • Quote: “Abstract Robotic manipulation requires models that generate executable actions while anticipating and evaluating their future consequences before physical execution. We present -World Model ( -WM), a unified video-action world model that integrates policy learning, video prediction, and action evaluation within a single future-predictive framework. Built on a shared video diffusion backbone, -WM provides two complementary interfaces. First, a video action model jointly predicts future visual lat”
  • Authors: pengfei-zhou; shengcong-chen; di-chen; et al.

EA-DQ-CONTRA-READ-0004

  • Claim: 数据质量必须经闭环评估定义；对机器人世界模型来说，短期视觉真实感不如长程、动作忠实的 rollout 一致性更能决定其作为策略评估器的可靠性。
  • Stance: support | Confidence: direct
  • Paper: 2607.02642 GigaWorld-1: A Roadmap to Build World Models for Robot Policy Evaluation
  • Locator: Abstract (full-text section)
  • Evidence: 论文指出真实机器人策略评估受硬件和人工监督限制，是基础模型迭代瓶颈；WMBench 用真实 teleoperation 数据和匹配 policy rollouts 构造评估，并分析 7 个视频世界模型、4 种动作表示和 324,000 余次模拟 rollout。其结论强调 evaluator 质量由长程 action-faithful rollout consistency、可迁移物理先验、动作编码、记忆和评估导向 post-training 共同决定。
  • Quote: “Using WMBench, we analyze 7 video world models, 4 action representation schemes, and over 324,000 simulated policy rollouts paired with real robot executions, further enriching our analysis with large-scale community submissions from the CVPR 2026 GigaBrain Challenge, curated synthetic trajectories, and a training videos spanning more than 12,000 hours. Our experiments deliver three core insights: evaluator quality is dominated by long-horizon, action-faithful rollout consistency rather than sho”
  • Authors: gigaworld-team; angyuan-ma; boyuan-wang; et al.

EA-DQ-CONTRA-READ-0005

  • Claim: 软物体和形变场景中的数据质量主要受可观测性约束：必须同时记录多视角全局运动、触觉局部形变和可用于评测的 3D 轨迹。
  • Stance: support | Confidence: direct
  • Paper: 2607.05390 Deform360: A Massive Multi-view Visuotactile Dataset for Deformable World Models
  • Locator: Abstract (full-text section)
  • Evidence: 论文认为形变物体有高维状态和复杂材料属性，接触诱发的局部形变常被末端执行器或物体遮挡；已有数据集常缺对象多样性、依赖合成数据，或缺高保真标注与接触形变。Deform360 采集 198 个日常物体、1,980 个交互序列、215 小时以上数据、41 个环视相机和双臂触觉 UMI gripper，并用 markerless 3D tracking 提取稠密几何与运动。
  • Quote: “To address this, we present Deform360, a large-scale visuotactile dataset featuring 198 daily-life objects, 1,980 interaction sequences, and over 215 hours of observations from 41 surround-view cameras and bimanual tactile grippers to capture both global motion and contact-induced local deformations. Leveraging a novel markerless visuotactile 3D tracking pipeline to extract dense geometry and motion, we systematically evaluate current state-of-the-art world models, comparing 2D video models agai”
  • Authors: hongyu-li; wanjia-fu; xiaoyan-cong; et al.

EA-DQ-CONTRA-READ-0006

  • Claim: Lift3D-VLA 指出，纯 2D VLA 难以保真地表达可达性、遮挡、接触和随时间演化的几何约束，而现有 2D‑3D 转换又会损失几何保真度。
  • Stance: support | Confidence: direct
  • Paper: 2607.06564 Lift3D-VLA: Lifting VLA Models to 3D Geometry and Dynamics-Aware Manipulation
  • Locator: I Introduction
  • Evidence: 引言将操作需求归结为显式 3D 结构与时间一致性，并说明纯 2D 管线及有损的跨模态变换会削弱这些约束。
  • Quote: “Despite this progress, robotic manipulation fundamentally requires spatial reasoning in the physical world [ 57 , 82 , 12 , 13 , 59 ] : the robot must infer 3D structure, reason about geometric relationships (e.g., reachability, occlusion, and contact), and plan actions that remain temporally consistent as the geometry evolves. Purely 2D VLA pipelines often struggle to reliably capture these geometric constraints, particularly in cluttered or dynamic environments. A natural direction is to expli”
  • Authors: jiaming-liu; qingpo-wuwu; nuowei-han; et al.

EA-DQ-CONTRA-READ-0013

  • Claim: VR 示教质量依赖交互模态和视觉表示，并且不同任务会偏好不同输入配置；采集系统优化不能只追求沉浸感或视觉保真。
  • Stance: conditional | Confidence: direct
  • Paper: 2602.10618 From Interaction to Demonstration Quality in Virtual Reality: Effects of Interaction Modality and Visual Representation on Everyday Tasks
  • Locator: 1 Introduction
  • Evidence: 论文指出 VR 用于记录机器人学习示教时，visual fidelity 可能不如 user behavior 的 quality/reliability 重要；输入设备与可视化会影响工作负荷、运动效率、不必要动作和执行精度。实验发现 controller 与 motion-capture gloves 在 pick-and-place 与 manner-oriented tasks 上呈现不同轨迹策略和权衡。
  • Quote: “In contrast, when VR is used to record demonstrations for robot learning, visual fidelity may be less important than the quality and reliability of user behavior during task execution.”
  • Authors: robin-beierling; manuel-scheibl; jonas-dech; et al.

EA-DQ-CONTRA-READ-0014

  • Claim: 人类视频能扩大机器人学习数据来源，但其质量取决于机器人可执行性和任务兼容性；仿真过滤可把不可达、估计错误或 grasp 不兼容的轨迹排除或标注出来。
  • Stance: conditional | Confidence: direct
  • Paper: 2602.13197 Imitating What Works: Simulation-Filtered Modular Policy Learning from Human Videos
  • Locator: 3.3 Trajectory and Grasp Filtering via Simulation
  • Evidence: PSI 将人类演示转换为 6DoF object pose trajectories 后在仿真中执行，用于过滤不适合机器人学习的数据；不适合原因包括 pose estimation errors 和机器人 physically unachievable trajectories，并生成 grasp suitability labels 以学习 task-oriented grasping。
  • Quote: “Now that we have converted the human demonstrations into 6 DoF object pose trajectories, the next step is to execute them on a robot in simulation. This serves two purposes. One is to filter out those that may not be suitable for robot learning. There are two main reasons a trajectory may be unsuitable. First, pose estimation errors can lead to inaccurate trajectories. Second, the extracted trajectory may not be physically achievable by the robot. In either case, it would be harmful to train the”
  • Authors: albert-j-zhai; kuo-hao-zeng; jiasen-lu; et al.

EA-DQ-CONTRA-READ-0009

  • Claim: Synthetic world-model data for robot learning needs embodiment anchoring: rendered robot motion plus explicit scene and object priors can synthesize demonstrations in novel objects, scenes, and viewpoints while reducing teleoperation burden.
  • Stance: conditional | Confidence: direct
  • Paper: 2606.02577 RoboDream: Compositional World Models for Scalable Robot Data Synthesis
  • Locator: Abstract (full-text section)
  • Evidence: RoboDream anchors generation to rendered robot motion, conditions on scene/object priors, and introduces retrieval-and-rebirth plus prop-free teleoperation to generate demonstrations and reduce real data collection cost.
  • Quote: “Abstract Scaling robot learning requires large-scale, diverse demonstrations, yet real-world data collection via teleoperation remains prohibitively expensive and time-consuming. While video diffusion models offer a promising avenue for data scaling, existing generative approaches are often limited to superficial visual augmentation, or suffer from embodiment hallucinations that yield physically infeasible motions. We present a generalizable embodiment-centric world model that achieves scalable”
  • Authors: junjie-ye; rong-xue; basile-van-hoorick; et al.

EA-DQ-CONTRA-READ-0010

  • Claim: Paired task-execution videos are useful but costly; self-distillation can partially replace curated task-video supervision by using unlabeled scene images, VLM-generated tasks and solutions, and VLM feedback as weak verification.
  • Stance: conditional | Confidence: direct
  • Paper: 2606.12072 World Model Self-Distillation: Training World Models to Solve General Tasks
  • Locator: Abstract (full-text section)
  • Evidence: WMSD frames supervised fine-tuning on paired task-execution videos as costly, then proposes self-distillation and reinforcement learning where a VLM generates tasks/solutions from unlabeled scene images and feedback verifies sampled videos.
  • Quote: “Abstract Pretrained video generators are promising visual world models that exhibit emergent task-solving abilities; however, their reliance on detailed textual descriptions limits their direct use for planning and decision-making. Existing approaches either outsource this reasoning to language or vision-language models, or rely on supervised fine-tuning with paired task-execution videos, which are costly to collect and difficult to scale. We propose a scalable framework that elicits task-solvin”
  • Authors: sebastian-stapf; pablo-acuaviva-huertos; aram-davtyan; et al.

EA-DQ-CONTRA-READ-0015

  • Claim: 混合质量的长程示教不应粗粒度整条丢弃；次优 episode 里可能含有高价值恢复片段，质量控制需要下探到 frame/chunk 级别的进展信号。
  • Stance: conditional | Confidence: direct
  • Paper: 2606.28320 WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation
  • Locator: Abstract (full-text section)
  • Evidence: 论文指出长程遥操作包含 pauses、fumbles 和 recoveries，整条 episode 过滤会丢失 otherwise suboptimal executions 中嵌入的 high-advantage segments，也无法剪掉保留示教中的局部 hesitation；WARP-RM 学习 dense relative progress 并用 WARP-BC upweight high-advantage action chunks。
  • Quote: “Abstract Scaling imitation learning requires large datasets, yet human teleoperation inevitably produces mixed-quality demonstrations containing hesitations and recoveries. Prior frame-level progress reward models supervise on absolute temporal progress proxies that suffer from label noise, or require costly human annotations to define subtask boundaries. We present WARP (Warp-Augmented Relative Progress), a novel fully self-supervised algorithm for learning dense, signed relative progress magni”
  • Authors: justin-yu; andrew-goldberg; kavish-kondap; et al.

EA-DQ-CONTRA-READ-0012

  • Claim: RynnWorld-Teleop将数字遥操作作为生成式数据引擎，但论文明确限定了它对精细流体动力学、高形变物体和跨机器人平台扩展的能力。
  • Stance: conditional | Confidence: direct
  • Paper: 2607.06558 RynnWorld-Teleop: An Action-Conditioned World Model for Digital Teleoperation
  • Locator: 6 Conclusion
  • Evidence: 结论的限制段指出，模型在精细流体和高形变操作上仍会失败，而当前跨本体迁移仍要求每个平台单独微调。
  • Quote: “Limitation. While RynnWorld-Teleop successfully demonstrates digital teleoperation as a viable data engine, several limitations remain. First, while the depth-modulated rendering captures 3D spatial dynamics, the model occasionally struggles with complex physical phenomena such as fine-grained liquid dynamics or the manipulation of highly deformable objects. Addressing these cases will likely require richer training data covering such interactions. Second, bridging the embodiment gap currently r”
  • Authors: haoyu-zhao; xingyue-zhao; hangyu-li; et al.

EA-DQ-CONTRA-READ-0011

  • Claim: Static image-text pretraining is insufficient training signal for contact-rich manipulation; world-model data should expose action-conditioned scene evolution and contact dynamics.
  • Stance: limit | Confidence: direct
  • Paper: 2606.12403 World Pilot: Steering Vision-Language-Action Models with World-Action Priors
  • Locator: Abstract (full-text section)
  • Evidence: World Pilot argues that VLA semantic grounding from static image-text pairs cannot capture continuous contact-rich dynamics, and uses WAM-derived scene-evolution and trajectory priors to complement the policy.
  • Quote: “Abstract Vision-Language-Action (VLA) models inherit semantic grounding from large-scale pretraining and perform competently across in-distribution manipulation tasks. This grounding, however, is built on static image-text pairs, whereas manipulation is a continuous, contact-rich process whose dynamics this pretraining cannot capture. We present World Pilot, a VLA framework that augments the policy with priors from a World-Action Model (WAM), routed into the decision chain through two complement”
  • Authors: zefu-lin; rongxu-cui; junjia-xu; et al.

EA-DQ-CONTRA-READ-0001

  • Claim: TACO 用触觉感知世界模型联合预测未来视频和力序列，再将真实失败附近状态转成带纠正动作的想象片段，用于接触丰富任务的 VLA 后训练。
  • Stance: limit | Confidence: direct
  • Paper: 2607.02840 TACO: TActile World Model as a Self-COrrector forScalable VLA Post-Training
  • Locator: 5 Conclusion and Limitations
  • Evidence: 结论的 Recognize–Imagine–Label 回路明确连接了真实失败、视频—力联合想象与纠正动作标注。
  • Quote: “We presented TACO, a tactile-aware world-model-driven framework for scalable VLA post-training in contact-rich manipulation. Following a Recognize–Imagine–Label loop, TACO converts real-world failures into imagined corrections without repeated human intervention: a tactile-aware world model jointly denoises future video and force sequences, while a unified progress-action model recognizes failure-adjacent states and labels imagined segments with corrective actions. To incorporate this supervisio”
  • Authors: shengbang-liu; yueru-jia; yuyang-yan; et al.

EA-DQ-CONTRA-READ-0002

  • Claim: SIEVE 按可复用原语组合和转换接口分配选择预算，再优先保留各组合模式中稳定、中心的轨迹；论文报告其用 50% 示教和 50% 训练步数可优于全量训练。
  • Stance: limit | Confidence: direct
  • Paper: 2607.06442 SIEVE: Structure-Aware Data Selection for Imitation Learning with VLA Models
  • Locator: Introduction
  • Evidence: 引言的贡献列表同时说明了结构暴露、学习友好轨迹选择和半量数据超过全量训练的结果。
  • Quote: “Our contributions are as follows: • We propose a primitive-compositional view of trajectory utility, realized by Primitive Discovery and Structural Exposure Allocation, which allocate selection budgets according to reuse-aware primitive and transition exposure under diminishing returns. • We introduce Learning-Friendly Trajectory Selection, which selects medoid trajectories within each composition-pattern bucket to favor central, stable, and predictable realizations for behavior cloning. • We pr”
  • Authors: changti-wu; bin-yu; zhaolong-shen; et al.

References

Interactive Wiki

进入交互阅读

静态页采用默认版本；v0、v0、v0可在同一交互 Wiki 中阅读，不建立相互竞争的索引页。

用 AI 继续研究 v0 v0 v0

Method and limits

研究方法、证据边界与引用说明

本页由当前结算的证据层自动派生，不改写已结算 run。论文只有在完整非 OCR 全文、结构化论文笔记和 claim-support audit 通过后，才进入接受证据。

页面结论是对所列论文与研究窗口的综合，不等同于无限外推。引用本专题时，请同时保留本页 canonical 地址、访问日期，并在需要时回溯原论文。