DeformSmith builds deformable objects for robot hands from one image
single source· 1 articles · confidence: medium · first seen 2026-09-16 20:00 UTC
What this means for you
Nothing to act on yet. The input is an abstract and a project page: no scores, no evaluation date, no released code. If you generate soft-body assets for manipulation training, the three baselines now to compare against are PhysGen3D, PhysGM and PhysX-Omni — wait for numbers before switching.
Posted to arXiv on 16 September 2026, DeformSmith is a system that builds simulated deformable objects — things that bend, squash or stretch when touched — for robot manipulation, starting from a text description or a single image. It generates geometry, material behaviour and robot interaction in stages, checking and refining each stage against a shared physics-based test loop. The authors report it beats PhysGen3D, PhysGM and PhysX-Omni on visual quality and physical plausibility, but the paper gives no scores, no evaluation dates and no numbers.
Key facts
- ·The paper is arXiv 2609.18620, posted 16 September 2026. source
- ·DeformSmith generates deformable objects for robot manipulation from text or a single image. source
- ·It builds geometry, physical models, material behaviour and robot interaction in stages, using a shared physics-grounded harness. source
- ·The authors report better visual quality and physical plausibility than the baselines PhysGen3D, PhysGM and PhysX-Omni. source
- ·The abstract gives no scores, no evaluation dates and no quantitative comparison against those baselines. source
- ·A project page is listed at https://can-lee.github.io/deformsmith-web/. source
What the sources say
- Hugging Face Daily Papers (research) — Abstract for a system that builds simulated soft objects for robot training, with no numbers attached.
Sources
The original reporting. Follow these — they did the work.
- Hugging Face Daily PapersDeformSmith: Physics Harness-Guided Hierarchical Generation of Deformable Assets for Robot Manipulation2026-09-16