跳到正文
arXiv cs.AI· Akul Santhosh, Chaitra Sathe, Aastha Jhunjhunwala·· 2 小时前AI 评分41

FailureGen:用 Cosmos 世界模型可扩展生成机器人失败数据

FailureGen: Scalable Robot Failure Data Generation with World Models

AI 导读

FailureGen 是一种通过复用已记录失败的动作轨迹、用世界模型生成带时序事件标注的合成失败视频的方法,包含微调动作条件 Cosmos 世界模型、标注、运动迁移和过滤四个阶段。在初步对比中,用真实加生成数据训练的失败检测器平均精度达 0.51(仅用真实数据为 0.41),36 个失败区间检出 33 个(真实数据为 28 个)。该方法可扩充机器人失败研究的数据供给。

正文 · 原文

View PDF HTML (experimental)

Abstract:Robot failures are difficult to collect at scale, limiting the range of failure scenarios represented in robot learning datasets. We present FailureGen, a method for generating synthetic videos with temporal event annotations by reusing action trajectories from recorded failures. The method comprises four stages: (1) fine-tuning an action-conditioned Cosmos world model on successful and failed real executions; (2) annotating source recordings to identify failure-associated donor segments and compatible successful hosts; (3) transferring donor motion into host contexts, with optional spatial offsets, to generate continuous video rollouts; and (4) annotating and filtering the generated samples according to observed events and visual quality. We study two implementations with different tasks and robot configurations, with visual failure detection serving as one downstream application. In a preliminary comparison, a detector trained on real and generated data achieves an average precision of 0.51, compared with 0.41 using real data alone, and detects 33 of 36 failure intervals, compared with 28 of 36. By reusing observed failures in additional contexts, FailureGen expands the data available for studying robot failures and developing downstream learning methods.
Subjects: Robotics (cs.RO); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.22367 [cs.RO]
  (or arXiv:2611.22367v1 [cs.RO] for this version)
  https://doi.org/10.48550/arXiv.2611.22367

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Akul Santhosh [view email]
[v1] Tue, 17 Nov 2026 08:21:16 UTC (13,180 KB)

来源:arXiv cs.AI · arxiv.org