Nebius Token Factory
Mini toolsDreamDojo
By DreamDojo
DreamDojo is a foundation world model that learns physical interaction by watching large-scale first-person human video, then distills that knowledge into robot-ready policies. Pretrained on the DreamDojo-HV dataset with 44k hours of egocentric footage, it generalizes across objects, scenes, and skills, then is post-trained for specific robot embodiments. After distillation, the model supports interactive rollouts at around 10 FPS for over a minute, making closed-loop planning and teleoperation feasible in simulation before deployment to real hardware.
New
Video
Gen 4
Overview
DreamDojo is NVIDIA’s generalist robot world model trained on 44k hours of egocentric human video, enabling real-time, action-conditioned simulation and planning for diverse robot bodies.
Tools using DreamDojo
No tools found for this model yet.
Last updated: February 12, 2026
