Function-Preserving Shape Deformation Enables Zero-Shot Real-to-Sim-to-Real Robot Manipulation
zhenjun_zhao · x · 2026-09-17
A paper introduces a function-preserving Real-to-Sim-to-Real framework that generates synthetic demonstrations from reconstructed assets without teleoperated trajectories, enabling zero-shot robot policy deployment.
- Problem: Contact-rich manipulation depends on precise geometric interfaces; standard shape augmentation distorts task-critical interfaces (fit mismatches, interpenetration), making downstream interaction infeasible.
- Method: Constraint-guided mesh deformation augments task-relevant object geometries, with physically consistent transfer of task poses and collision proxies; visual domain randomization is applied during simulation rollouts.
- Results: Policies achieve robust zero-shot generalization across unseen object geometries and diverse visual conditions in contact-rich, long-horizon tasks, without real-world fine-tuning.
The approach offers a practical path to scalable robot learning via shape deformation.
More from Embodied
- GPT-Policy: In-Context Robot Learning with VLM Agents, No Gradient Updates — Dongzhou Cheng · 2026-09-17
- World Labs' Atlas Scans by Generative Guessing; NeRF Creator Admits Productization Is Hard — cen6wkf · 2026-09-17
- CXMT's LPDDR5X lands in flagship phone as Nubia ships $885 Doubao AI handset — pstAsiatech · 2026-09-17
- Key open challenges for VLAs: language, evaluation, deployment, causal reasoning — abursuc · 2026-09-17
- LADA: latent actions imitate language from few observation-language pairs — abursuc · 2026-09-17
- A Primer on Latent Action Models From the #ssad2026 Talks — abursuc · 2026-09-17