SkeLAM grounds latent actions in robot skeletons for controllable world models
Qi Lv · hf · 2026-10-23
Researchers from HIT Shenzhen, NUS Show Lab, and Ping An Tech propose SkeLAM, a skeleton-guided latent action model for controllable robot world modeling.
- Problem: reconstruction-driven latent actions conflate robot motion with object responses and scene changes, breaking correspondence with robot controls.
- Method: decomposes latent actions into a kinematic component grounded by robot skeleton supervision and a residual component retaining complementary visual dynamics; cross-modal swap reconstruction makes RGB-derived and skeleton-derived kinematic representations interchangeable.
- Result: the world model conditions on both via separate pathways with branch dropout, and a lightweight adapter maps robot actions to the kinematic space, enabling controllable future prediction and motion transfer after fine-tuning.
More from Embodied
- Chinese robotics companies' overseas sales heads often don't speak English, observer says — broodsugar · 2026-10-23
- Asimov ships new grippers and an extra arm DOF with third-party support — freelerobot · 2026-10-23
- NAVER AI Lab releases LaLa-World, scaling latent world models with natural-language actions — naver-ai · 2026-10-23
- Ken Goldberg: humanoid lab success rates stalled at 70-80%, 'not acceptable for real industry' — AIFlow_ML · 2026-10-23
- Inside Li Zexiang's InnoX Academy: seed rounds paid in workers and factory space — DominiqueCAPaul · 2026-10-23
- Optimus Is in Production but Its Hands Aren't Ready: Inside the Humanoid Bottleneck — AIFlow_ML · 2026-10-23