DreamDojo: A Generalist Robot World Model from Large-Scale Human Videos
Original research
MODEL PROFILE · Enabling world model
DreamDojo
Learns action-controllable robot-world simulations from large-scale human video.
Research organisationGao and collaborators
Research approachLatent-action modelling
Published year2026
BackboneRobot world-model video backbone
Overview
DreamDojo is a foundation robot world model that uses latent actions to make simulations controllable. Its primary emphasis is world simulation for applications including planning and evaluation: it must not be confused with a standalone robot policy.
Source record
Core research identity and broad method were source-checked as of 2026-11-03. Independent benchmarking, commercial availability, weights and licensing have not been assessed.