跳到正文
返回论文列表
cs.CV提交于 已译

看似合理的视频可能导致糟糕决策:诊断与 CURE 世界模型用于规划

Plausible Videos Can Lead to Poor Decisions: Diagnosing and CURE-ing World Models for Planning

Raphael C. Kim · Michele Santacatterina · Jingsen Zhu · Ramin Zabih

中文摘要

动作条件世界模型(action-conditioned world models)越来越多地被用作规划的世界模型,然而,一个能生成与训练数据一致、看似合理且逼真的视频的世界模型,却可能产生糟糕的决策。我们探讨两个关键问题:(1)何时可以信任世界模型进行规划;(2)当无法信任时应该怎么做。我们提出 CURE(Certify, Upweight, Rollout, Explore),一种证书引导的技术,用于评估世界模型是否可被信任用于规划,并决定下一步行动:是上调训练数据权重以去偏模型;在测试时推出更多样本;还是在支持度(训练数据对规划器决策的覆盖程度)缺失的地方通过收集新数据进行探索。我们在多个领域部署 CURE,涵盖基于 PhyWorld 的模拟物理、MuJoCo 机械臂控制,以及模拟和真实视频中的 Language-Table 推演。我们的证书所信任的决策比不信任的决策具有更低遗憾(例如,在 Language-Table 模拟中为 0.06 对 0.38,在 MuJoCo 接触中为 0.07 对 0.38),而视频感知质量并不能识别出好的决策。通过关注决策质量而非感知质量,CURE 在相同数据下,在不同支持度水平上比最佳悲观决策规则降低遗憾 10-25%,比世界-动作模型、搜索和策略学习降低 41-64%。

关键要点

  1. 01问题:看似合理的视频世界模型可能产生糟糕规划决策,感知质量无法识别决策好坏。
  2. 02方法:提出 CURE 证书引导技术,评估世界模型可信度并决定上调权重、推出或探索。
  3. 03结果:证书信任的决策遗憾更低(如 0.06 vs 0.38),比基线降低遗憾 10-64%。
  4. 04局限:未明确提及。

解读

尚无解读。

原始英文摘要

arXiv:2611.22561v1 Announce Type: new Abstract: Action-conditioned world models are increasingly used as world models for planning, yet a world model that makes plausible, realistic-looking videos that are consistent with the training data can produce poor decisions. We address two key questions: (1) when can we trust a world model for planning, and (2) what should we do when we cannot? We introduce CURE (Certify, Upweight, Rollout, Explore), a certificate-guided technique that assesses whether a world model can be trusted for planning and determines what to do next: whether to upweight the training data to debias the model; roll out more samples at test time; or explore by collecting new data where support (how well the training data cover the planner's decisions) is missing. We deploy CURE across multiple domains, spanning simulated physics based on PhyWorld, MuJoCo arm control, and Language-Table pushing in simulation and on real videos. The decisions our certificate trusts have lower regret than those it does not (e.g., 0.06 vs. 0.38 on Language-Table sim and 0.07 vs. 0.38 on MuJoCo contact), whereas video perceptual quality does not identify good decisions. By focusing on decision quality rather than perceptual quality, CURE lowers regret over the best pessimistic decision rule by 10-25% across support levels, and over world-action models, search and policy learning by 41-64%, given the same data.

同方向论文 · cs.CV

查看全部 →

反向发现 · 这篇论文相关的工具与提示词

去交叉发现 →

关联工具 · 1

NVIDIA Blog72

NVIDIA Cosmos 扩展世界模型训练、评测与部署工具

NVIDIA Cosmos 新增训练工具、仿真检查点与 Cosmos Data Factory 蓝图,用于构建训练数据、训练和适配世界模型,并完成仿真评测与部署。开发者可微调 GR00T、Alpamayo、QWEN、Pi0.5 等模型;Cosmos 3 NIM 在已验证配置中可实现最高 2 倍推理提速,Metropolis Blueprint 3.3 可在同一 GPU 上监控多 40% 的视频流。

推荐理由:新增工具覆盖训练数据构建、模型训练优化、策略测试与部署,并提供了具体应用效果。