Event
RynnWorld-Latent Cross-Embodiment World Model Released
2 reports · 2 sources
Haoyu Zhao and other authors released RynnWorld-Latent, a cross-embodiment video world model that uses the 3D-grounded latent actions proposed by RynnLAM as a shared conditioning interface, enabling joint pretraining of human interaction videos, robot demonstrations, and simulation data without the need for a unified native action space. The authors state that the model was pretrained on 23 datasets, approximately 553.5 million frames, and is adapted to target robots through post-training. The authors also state that, under their experimental setup, mid-training on synthetic data improved policy success rates on LIBERO and LIBERO-Plus, and that both generation routes also improved real-world policy success rates compared with training only on real demonstrations.
Report timeline
- RynnWorld-Latent: A Cross-Embodiment World Model Conditioned on Universal Latent Actions
RynnWorld-Latent is a cross-embodiment video world model that uses the 3D-grounded latent actions introduced by RynnLAM as a shared conditioning interface, enabling joint pretraining on human interaction videos, robot demonstrations, and simulation data without requiring a unified native action space.
arXiv cs.RO ·
- RynnWorld-Latent: A Cross-Embodiment World Model Conditioned on Universal Latent Actions
The research team proposes RynnWorld-Latent, a cross-embodiment video world model that uses the 3D-grounded latent actions of RynnLAM as a shared conditioning interface, enabling joint pretraining on human interaction videos, robot demonstrations, and simulation data.
HuggingFace Daily Papers ·