Uranus: Building the Next-Generation Simulation Infrastructure for Embodied AI
Uranus generates open-ended robot simulation rollouts at 24 FPS from future joint-position trajectories.
The paper describes a data-driven simulator built around a joint-trajectory-conditioned autoregressive diffusion model. It generates one latent frame per step, corresponding to four RGB frames, and supports synchronized multi-view output across different robot bodies and camera setups. The authors say they evaluated it on in-distribution and out-of-distribution data, while also identifying current limitations. They also released code, data, SDK materials, and model weights. HF Daily Papers' note
The paper describes a data-driven simulator built around a joint-trajectory-conditioned autoregressive diffusion model. It generates one latent frame per step, corresponding to four RGB frames, and supports synchronized multi-view output across different robot bodies and camera setups. The authors say they evaluated it on in-distribution and out-of-distribution data, while also identifying current limitations. They also released code, data, SDK materials, and model weights. HF Daily Papers' note
score 5