Robotics · 2026-10-05
Awomo-SimDataEngine: Agentic Simulation-ReadyWorld Generation
PhysicalRSI Team, Danjiao Ma, Enhui Ma, Haohan Liu, Heng Jia, Hui Shan, Jianhua Xu, Jiahuan Zhang, Jiangdi Xu, Kaiwen Guo, Kaicheng Yu, Linwei Zhang, Liyang Jin, Maochun Luo, Pengyao Niu, Shiwen Li, Shuangyu Feng, Tong Zhang, Tianheng Wang, Xin Wang, Xiangru Huang, Yongqiang Huang, Zhaozhi Wang, Zijian Ma
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
Awomo-SimDataEngine is an agentic system for generating simulation-ready robot worlds and synthesizing replayable demonstrations. It combines ISArt for articulated asset generation, Unravel for single-image scene reconstruction, and SimForge for text-driven single- and multi-room scenes, coordinated by a graph-native harness with validation and bounded repair plus a PolicyForge interface for demonstration synthesis. On MuJoCo-based LIBERO-Plus, SimData-Ultra improves WAM success from 77.17% to 89.43%.
Why it matters
The authors argue that robot-training data needs physically valid interaction-ready worlds and repeatable task execution, not just plausible appearance. Awomo-SimDataEngine integrates asset/scene construction, validation and bounded repair, and demo synthesis so that generated worlds can be accepted and used for policy co-training. The reported LIBERO-Plus results provide evidence that the generated data can improve downstream world-action model performance across simulator boundaries, with the largest improvement in goal generalization.
Method
- ISArt generates articulated object assets as textured part meshes plus a URDF kinematic tree with joint origins/axes/ranges, followed by Isaac Sim validation and repair-driven retries.
- Unravel reconstructs an editable interactive scene from a single-view image via verified layer-wise deconstruction (object removal + point-map fusion) and reverse-order recomposition with support-aware settling.
- A Graph-Native Harness coordinates module execution, validation evidence, and bounded repair with owner routing; PolicyForge binds accepted worlds to task/robot/reset contracts to synthesize replayable demonstrations in Isaac Sim.
Abstract (from arXiv)
Generating useful robot-training data requires more than visually plausiblescenes: objects must support interaction, placements must remain physicallyvalid, and tasks must admit repeatable execution. We present\textbf{Awomo-SimDataEngine}, an agentic system that connects asset and scenegeneration to robot demonstration synthesis. Shared asset services providerigid and articulated objects, including structure-grounded part and jointgeneration with ISArt. Scene generation supports two complementary routes:Unravel reconstructs editable scenes from images, while SimForge buildssingle-room and multi-room environments from text. A graph-native harnesscoordinates construction, validation, andbounded repair, routing failures to the responsible module while retainingunaffected scene state. PolicyForge binds validated worlds to tasks and robotembodiments to produce replayable demonstrations. Evaluations cover assetgeometry, scene quality, and downstream policy learning. On MuJoCo-basedLIBERO-Plus, co-training with Isaac Sim demonstrations improves the overallsuccess rate of a World-Action Model (WAM) from $77.17\%$ to $89.43\%$. Goal and spatialsuccess improve by $31.66$ and $6.25$ percentage points, respectively.These results support the utility of the generated data for cross-simulatorpolicy training, with more limited gains on long-horizon tasks.
Related papers
- EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation
- Real2Gym: Building Gyms from Videos, Bringing Skills to Robots
- AffordCraft: Scalable Construction of Task-Ready Simulation Assets from Single Images
- Code2Worlds-2: Physically Plausible Agent Harness for Embodied World Simulation
- SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-01 · Indexed by Pub-AI: 2026-10-05 UTC
Awomo-SimDataEngine: Agentic Simulation-ReadyWorld Generation
Preprint posted: 2026-10-01 · Original record ↗
- arxiv
- 2610.02274
Explicit source versions (1)
- v1 · 2026-10-01
- Preprint available
Source date: 2026-10-01 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Generating useful robot-training data requires more than visually plausiblescenes: objects must support interaction, placements must remain physicallyvalid, and tasks must admit repeatable execution. We present\textbf{Awomo-SimDataEngine}, an agentic system that connects asset and scenegeneration to robot demonstration synthesis. Shared asset services providerigid and articulated objects, including structure-grounded part and jointgeneration with ISArt. Scene generation supports two complementary routes:Unravel reconstructs editable scenes from images, while SimForge buildssingle-room and multi-room environments from text. A graph-native harnesscoordinates construction, validation, andbounded repair, routing failures to the responsible module while retainingunaffected scene state. PolicyForge binds validated worlds to tasks and robotembodiments to produce replayable demonstrations. Evaluations cover assetgeometry, scene quality, and downstream policy learning. On MuJoCo-basedLIBERO-Plus, co-training with Isaac Sim demonstrations improves the overallsuccess rate of a World-Action Model (WAM) from $77.17\%$ to $89.43\%$. Goal and spatialsuccess improve by $31.66$ and $6.25$ percentage points, respectively.These results support the utility of the generated data for cross-simulatorpolicy training, with more limited gains on long-horizon tasks.