0: A Generalist Policy for Multi-Embodiment Motion Control

We’re building a foundation motion policy for any robot, on the ground or in the air, with wheels, legs or rotors.

It learns across millions of embodiments derived from a growing collection of more than 200 realistic robot models. With every robot added, the policy becomes more capable and better able to transfer to new robots.

Contribute your robot and help us build the largest open-source robot collection for multi-embodiment training and deployment!

Open the playground Contribute your robot

Real-world deployments

Unitree Go2 · Silver Badger · Unitree G1

TU Darmstadt

Booster K1

University of Amsterdam

Unitree H2

University of Southern California

SpaceR Floating Platform

University of Luxembourg

Unitree B1

UMass Amherst

ANYmal D

ETH Zurich

Quadrotor

University of Zurich

BlueROV2 Heavy

Heriot-Watt University

Reach Alpha 5

Heriot-Watt University

Hexbug

NTNU

BlueROV2 Heavy

NTNU

Rover

University of Luxembourg

Rover

University of Luxembourg

Unitree R1

Montanuniversität Leoben

TIAGo Pro

Inria

Shelfy

Inria

TIAGo++

TU Darmstadt

Ulixarm

Itaca Robotics

Kinova Gen3

Cornell University

CHARM hand

ETH Zurich

Terra++ Lite 6

University of Illinois

TurtleBot 4

University of Technology Nuremberg

Unitree Aliengo

University of Trento

Unitree Go2 · Silver Badger · Unitree G1

TU Darmstadt

LimX Tron2 SF

LimX Dynamics

LimX Tron2 WF

LimX Dynamics

Sanity

University of Cambridge

Fauna Sprout

Princeton University

UBTECH Walker TienKung EI

IIT · CUHK

LimX Luna

LimX Dynamics

Unitree H1

TU Darmstadt

ModalAI Starling

Stanford University

LimX Tron2 DASF

LimX Dynamics

VinRobotics VR-M3

VinRobotics

ToddlerBot

Stanford University

LimX Tron2 DASF2

LimX Dynamics

RealMan RealBot-01

Stanford University

Dash

UMass Amherst

Unitree H1-2

LAAS-CNRS

LimX Oli

LimX Dynamics

TidyBot++

Princeton University

Training methodology

We take Unified Robot Morphology Architecture (URMA) and Embodiment Scaling Laws to the extreme by randomizing robot embodiments on the fly throughout training. This exposes a single policy, conditioned on each embodiment, to the broadest possible range of robot morphologies. By learning from this diversity, the policy develops generalizable motion skills that transfer to new embodiments.

Our robot collection

6 of 540 robots

1X Neo

Adam Lite

Agibot A2

AgiBot A3 T2.5

Agibot D1

Agibot G1

Contribute to 0

Help us build a foundation motion policy that transfers across robots on the ground and in the air, with wheels, legs or rotors, supported by the largest open-source robot collection.

We have two ways for you to get involved!

The deadline for training and deployment collaborations was October 15, but we continue to accept new robot models at contribute@gamma-zero.com.

Contribute your robot model

Share a valid URDF for a real-world robot that we can simulate with MuJoCo, including meshes, inertial parameters, joint limits, and actuator torque and velocity limits. Accepted models will become part of the open-source collection, and contributors will be cited or acknowledged. We welcome legged robots, wheeled robots, manipulators and drones. Standalone robot hands are outside the scope of this collection.

Train and deploy with us

Collaborate with us to extend the training pipeline and deploy γ₀ on new hardware. Contributions spanning both training and real-world deployment may qualify for authorship.