Ropedia W0: cross-embodiment world model trained on 10k hours of human video hits SOTA
liuziwei7 · x · 2026-11-03
Ropedia W0 tackles the hardest part of robot world models: whether actions transfer across bodies. Instead of training separate models per embodiment, it uses camera poses (Plücker-ray maps) for locomotion and projected kinematic skeletons for manipulation, putting humans and robots in one shared representation.
- Data: Xp-Data with 10,000 hours of egocentric human video plus 3,000 hours of robot data, uniformly annotated; also introduces Xp-Teleop, a long-horizon benchmark built on humanoid teleoperation data.
- Results: SOTA across five cross-embodiment benchmarks (DreamDojo, Ψ0, ω-HOME, Xp-Teleop, AgiBot) and held-out human videos; distilled to run minute-scale rollouts in real time for VR interaction.
- Costs: human data outweighs robot data 3x; robot skeletons need per-embodiment forward kinematics and heavy manual keypoint annotation; real-time inference requires distillation.
More from Embodied
- YC-backed Cyclon ships AI phone OS on Pixel 8a, promising apps that act for you — Scobleizer · 2026-11-03
- YC-backed Subvocal builds RF smart glasses that decode silent speech into text — aarushyadav · 2026-11-03
- GPT-6 Astra tops robot index, but finishes dual-arm sim task only twice in 800 trials — minchoi · 2026-11-03
- Starcloud-1 marks one year in orbit with the only NVIDIA H100 in space, 3 more launches planned — ycombinator · 2026-11-03
- Ant Group's Robbyant shows LingBot-VLA 2.0: 60k hours of real data, runs on 20+ robots — huangyun_122 · 2026-11-03
- Delta Intelligence unveils Δ₀, a humanoid foundation model for whole-body loco-manipulation — chris_j_paxton · 2026-11-03