World models for robotics and driving
These action-conditioned models predict what a robot or a vehicle will see after it acts.
Teams use them to test and train control policies in simulation before anything moves in the real world.
10 models · Alphabetical, never ranked · Updated 2026-11-03
Robotics
1X World Model
1X Technologies
Robot action and video modeling
Permissive licenseOpen weightsDriving
comma.ai world model 4B
comma.ai
Frame history, ego motion and future plan anchors
Permissive licenseOpen weightsPhysical AI
Cosmos Predict2.5
NVIDIA
Text, image and video conditioning
Custom termsOpen weightsRobotics
Ctrl-World
Stanford University and Tsinghua University
Keyboard commands converted into robot action chunks
Permissive licenseOpen weightsRobotics
EnerVerse-AC
AgiBot
Robot actions and camera parameters
Non-commercialOpen weightsDriving
Oasis 3
Decart
Steering, movement and API control signals
Service termsNo public weightsReal-time claimTry on the official site
Physical AI
Odyssey-3
Odyssey
A prompt, then a person's or an agent's actions
Service termsNo public weightsReal-time claimTry on the official site
Driving
Orbis
University of Freiburg
Past driving video frames
Permissive licenseOpen weightsDriving
Orbis 2
University of Freiburg
Past driving video and an optional steering trajectory
Permissive licenseOpen weightsDriving
Vista
OpenDriveLab
Driving trajectory conditioning
Permissive licenseOpen weights
Every model listed here has its own page, a resolved license reading and at least one way to run or try it. Models that do not clear that bar stay in the full directory until their sources do.
The data is open and accepts corrections. View the dataset on GitHub