CMU's EgoX: cross-embodiment egocentric manipulation via "embodiment dreaming" pretraining
CarnegieMellonU · hf · 2026-11-14
Researchers from CMU, Bosch AI, NVIDIA, UC Berkeley and UT Arlington released EgoX, an egocentric cross-embodiment manipulation learning framework.
- Method: Pretrains a modular Transformer with an "embodiment dreaming" strategy on heterogeneous human, robot, and simulated demonstrations, then fine-tunes for target tasks.
- Results: Improves manipulation task progress, especially under out-of-distribution conditions, and extends to long-horizon loco-manipulation.
- Architecture uses a shared trunk with action experts and a dream expert, enabling fast adaptation to new embodiment-task domains.
Project page: ego-dream.github.io
More from Embodied
- Xiaomi robot completes 50+ task chains across 3 rooms in 30-min real-world trial — CurieuxExplorer · 2026-11-14
- CoRL spotlight GaP debuts as open-source robotics sim RSS hits 500+ GitHub stars — erwincoumans · 2026-11-14
- Radical AI's self-driving lab at Sandia Labs spins an AI flywheel for materials discovery — CatAstro_Piyush · 2026-11-14
- Smearing vaseline on a robot camera: 5 frontier models attempt dangerous tasks at 97% — 量子位 · 2026-11-14
- Two Robots Given Claude Haiku End Up Debating the Meaning of Life — freelerobot · 2026-11-14
- Open-Source Humanoid Robot Asimov Hits Day 382, Tests New Outdoor Walking Gait — freelerobot · 2026-11-14