arXiv：cs.LG· Juntao Ren, Yifan Hou, Shuran Song·· 3 小时前AI 评分38
NEEDLE：用经核验的局部拼接离线重写机器人数据
NEEDLEWORK: Offline Rewriting of Robot Data with Verified Local Stitches
AI 导读
NEEDLE 是一种离线数据集增强算法，仅用 RGB 图像、本体感知和回合级结果，在高维机器人演示的观测之间添加经核验的短动作桥接，无需新环境交互或特权物体状态。它通过采样技术将接受的桥接纳入策略训练，不合成中间图像也不丢弃原始演示。在真实机器人任务上，NEEDLE 相比各任务最强基线平均提升 21 个百分点成功率。
正文
Abstract:Robot demonstrations may contain useful behavior even when individual episodes are inefficient or unsuccessful. Trajectory stitching offers a way to compose these behaviors into improved training data, but identifying useful connections and verifying their feasibility is difficult in high-dimensional robot data, where many prior methods rely on low-dimensional state representations. We introduce NEEDLE, an offline dataset-augmentation algorithm that addresses these challenges by adding short, verified action bridges between recorded observations in high-dimensional robot demonstrations. First, NEEDLE identifies and creates connections that bypass suboptimal detours, broaden action coverage, and augment the original dataset with failed trajectories, using only RGB images, proprioception, and episode-level outcomes, without new environment interaction or privileged object state. Next, we present a sampling technique that incorporates accepted bridges into policy training without synthesizing intermediate images or discarding the original demonstrations, allowing policies to learn alternative actions while retaining the original dataset's coverage. On real-robot tasks, NEEDLE improves success rate over the strongest baseline on each task by an average of 21 percentage points. Videos and supplementary materials are on this https URL.
| Comments: | Submitted to ICLR 2027. Project page: this https URL |
| Subjects: | Robotics (cs.RO); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.02339 [cs.RO] |
| (or arXiv:2610.02339v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02339 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Juntao Ren [view email]
[v1]
Thu, 1 Oct 2026 18:11:32 UTC (7,986 KB)
来源：arXiv：cs.LG · arxiv.org