EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation
EmbodiedSmith generates diverse robot simulation data by recursively co-refining scenes and tasks for foundation-model training.
EmbodiedSmith is a simulation framework for scaling training data for robotic foundation models through recursive self-improvement. It jointly generates assets, scenes, and tasks, letting scene generation anticipate task needs and task generation drive scene edits. The pipeline covers mobile manipulators, humanoids, dexterous hands, deformable objects, and fluids. Experiments report higher task-generation success and better policy generalization from increased data diversity.
- Unifies asset, scene, and task generation in one simulation pipeline.
- Agentic loop co-refines scenes and tasks, including long-horizon tasks.
- Supports mobile manipulators, humanoids, dexterous hands, deformables, and fluids.
- More diverse generated data improved downstream policy generalization.
From the original article
1 min read at huggingface.co
Scaling robotic foundation models requires diverse training data and reliable evaluation environments. Simulation offers a scalable solution, yet existing generation pipelines remain constrained by predefined assets and skills, a disconnect between scene generation and task generation, and limited support for complex embodiments and physics.
By Yikai Qin, Yifei Deng, Mingjian Liang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.