Robotics · 2026-10-30

RoboAtom: Large-scale Atomic-Action Supervision for Robot Learning

Xueyuan Wei, Xuanxu Wang, Xudong Li, Yuping He, Wanshui Gan, Jiaqi Peng, Wenqi Guo, Jianjun Zhou, Junting Dong, Chunhua Shen, Haoyu Guo, Mulin Yu, Tao Lu

arXiv:2610.40207PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

RoboAtom builds RoboAtom-Data by adding unified, time-aligned atomic-action annotations (skill category, action description, start/end frames) to trajectories from 16 sources. It also introduces RoboAtom-Bench and trains RoboAtom-Anno for automatic atomic-action annotation and hierarchical execution for robot control, reporting improved annotation performance and task success over task-level baselines on multiple benchmarks.

Why it matters

Fine-grained, time-aligned atomic-action labels can expose intermediate trajectory structure for skill recognition, temporal boundary localization, and skill-level execution. The authors address cross-source inconsistency with a unified annotation protocol plus same-task boundary propagation (PhysExtend) and multi-stage quality control, and then standardize evaluation via RoboAtom-Bench for both annotation and downstream robot control.

Method

  • RoboAtom-Data defines each atomic action as a segment with a shared skill category, an action description, and start/end frames; new segments start when the dominant skill changes, absorbing preparation/transition into adjacent segments.
  • Human-in-the-loop seed annotation refines skill semantics and temporal boundaries; same-task extension uses PhysExtend to propagate internal boundaries using skill-aware physical-event cues with dynamic programming, correction, and monotonic projection.
  • RoboAtom-Anno is trained on RoboAtom-Data and evaluated on RoboAtom-Bench (skill recognition and temporal boundary localization) and WGO-Bench; for control, atomic-action supervision with hierarchical execution improves task success over task-level baselines.

Limitation

In quality checks, propagation errors mainly arise from differences in execution structure within a task, sensor representations across embodiments, and reference annotations; tasks where the reference and target trajectories no longer share the same skill sequence are excluded from annotation extension.

Abstract (from arXiv)

Fine-grained action annotations can support skill reuse and compositional generalization in robot policies. However, robot datasets typically provide task-level instructions without subtask annotations. Although some include subtask labels, inconsistent formats and conventions hinder cross-source joint training. We propose RoboAtom-Data, which adds unified, time-aligned atomic-action annotations to existing data from multiple sources, covering 1.45 million trajectories, 6.88 million skill-labeled segments, and 16,882.1 hours. Our pipeline combines a unified protocol, large-scale manual annotation, propagation across repeated executions with the same ordered skill sequence, and multi-stage quality control. We evaluate the dataset through automatic atomic-action annotation and robot control. We construct RoboAtom-Bench to standardize evaluation of skill recognition and temporal boundary localization. Trained on the annotated data, RoboAtom-Anno achieves better automatic annotation performance than all evaluated open- and closed-source models on both RoboAtom-Bench and the external WGO-Bench. For robot control, atomic-action supervision paired with hierarchical execution improves task success over task-level baselines on LIBERO-Long, LIBERO-Plus, and RoboTwin. We will release the data, benchmark, model, and construction pipeline to support further research on skill-level robot learning.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First announced by arXiv: 2026-10-30 · Indexed by Pub-AI: 2026-10-30 UTC

preprint

RoboAtom: Large-scale Atomic-Action Supervision for Robot Learning

Xueyuan Wei, Xuanxu Wang, Xudong Li, Yuping He, Wanshui Gan, Jiaqi Peng, Wenqi Guo, Jianjun Zhou, Junting Dong, Chunhua Shen, Haoyu Guo, Mulin Yu, Tao Lu

First announced by arXiv: 2026-10-30 · Original record ↗

arxiv
2610.40207
Explicit source versions (1)
  • v1 · Date not supplied
  1. Preprint available

    Source date: 2026-10-30 · First detected: 2026-10-30 UTC · Historical baseline

    Source evidence (1)
    • Official evidence ↗ · RSS guid=oai:arXiv.org:2610.40207v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv

Fine-grained action annotations can support skill reuse and compositional generalization in robot policies. However, robot datasets typically provide task-level instructions without subtask annotations. Although some include subtask labels, inconsistent formats and conventions hinder cross-source joint training. We propose RoboAtom-Data, which adds unified, time-aligned atomic-action annotations to existing data from multiple sources, covering 1.45 million trajectories, 6.88 million skill-labeled segments, and 16,882.1 hours. Our pipeline combines a unified protocol, large-scale manual annotation, propagation across repeated executions with the same ordered skill sequence, and multi-stage quality control. We evaluate the dataset through automatic atomic-action annotation and robot control. We construct RoboAtom-Bench to standardize evaluation of skill recognition and temporal boundary localization. Trained on the annotated data, RoboAtom-Anno achieves better automatic annotation performance than all evaluated open- and closed-source models on both RoboAtom-Bench and the external WGO-Bench. For robot control, atomic-action supervision paired with hierarchical execution improves task success over task-level baselines on LIBERO-Long, LIBERO-Plus, and RoboTwin. We will release the data, benchmark, model, and construction pipeline to support further research on skill-level robot learning.