Ropedia W0 models loco-manipulation across humans and robots
Original title: Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
After distillation, the authors say it supports real-time, minute-long rollouts and live VR interaction.
Researchers present Ropedia W0, an egocentric world model for locomotion and manipulation across humans and robots. Training uses Xp-Data, with 10k hours of human video and 3k hours of robot data under unified annotations. They report state-of-the-art results on five cross-embodiment benchmarks and held-out human video, from zero-shot inference through full post-training.
- Xp-Data has 10k hours of egocentric human data and 3k hours of robot data.
- Xp-Teleop is a long-horizon benchmark from dexterous humanoid teleoperation data.
- Named benchmarks are DreamDojo, Ψ_0, ω-HOME, Xp-Teleop, and AgiBot.
- Scaling human pre-training improved both human and robot evaluations.
From the original article
1 min read at huggingface.co
World models enable embodied agents to anticipate action consequences, yet jointly modeling locomotion and manipulation across diverse embodiments remains challenging due to heterogeneous action spaces and limited multimodal supervision. We present Ropedia W0, the first egocentric loco-manipulation world foundation model spanning humans and robots with diverse embodiments.
By Tianqi Liu, Hanfei Zhao, Qifeng Liang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.