Skip to content
← All builds
Muted demo loop from the projectCredit: unitreerobotics/unitree_rl_gym
Train a walking policy for Unitree Go2, H1 or G1 in sim: media published by unitreerobotics/unitree_rl_gym (image 2)
Train a walking policy for Unitree Go2, H1 or G1 in sim: media published by unitreerobotics/unitree_rl_gym (image 3)

Train a walking policy for Unitree Go2, H1 or G1 in sim

Quad, Hard

Unitree RL GYM trains reinforcement-learning locomotion in Isaac Gym, validates it in MuJoCo, then deploys it to physical Unitree robots.

View build on GitHub ↗Share on X

The official repo supports Go2, H1, H1_2 and G1 with a train, play, sim2sim (MuJoCo), sim2real flow. You can do the whole thing without owning the robot.

Hardware & cost: An NVIDIA GPU for Isaac Gym. A Unitree robot only if you deploy (Go2 starts in the low thousands of dollars).

Difficulty: Hard.

Agent install prompt (paste into Claude Code, Cursor or Codex): "Set up https://github.com/unitreerobotics/unitree_rl_gym following the README (Isaac Gym + rsl_rl), train the Go2 task for a short run, then play it back in MuJoCo sim2sim."

Source: unitreerobotics/unitree_rl_gym on GitHub — https://github.com/unitreerobotics/unitree_rl_gym (facts checked against the project README, Oct 2026; prices vary by region).

Quick answers

What do you need to train a walking policy for Unitree Go2, H1 or G1 in sim?
An NVIDIA GPU for Isaac Gym. A Unitree robot only if you deploy (Go2 starts in the low thousands of dollars).
How hard is it to train a walking policy for Unitree Go2, H1 or G1 in sim?
Hard.

How your agent drives it

Copied from the unitreerobotics/unitree_rl_gym README

  1. Shell
    $ python legged_gym/scripts/train.py --task=xxx
  2. Shell
    $ python legged_gym/scripts/play.py --task=xxx
  3. Shell
    $ python deploy/deploy_mujoco/deploy_mujoco.py g1.yaml

Any coding agent that can run a shell can run these. Read the README before you run anything on a robot that moves.

Ideas to try with it

Our ideas, not the builder's claims.

  1. Train a walking policy in Isaac Gym and check it in MuJoCo before touching hardware
  2. Teach a Go2 a slower, quieter gait for indoor use
  3. Compare two reward settings and keep the one that falls less
1 more idea
  1. Have your agent run a training sweep overnight and report the best checkpoint
Source
unitreerobotics/unitree_rl_gym on GitHub
Last commit
2025-07-25
Builder
unitreerobotics
Repo
unitreerobotics/unitree_rl_gym, 3,654 stars
Built with
Unitree Go2, Unitree G1, Unitree H1

Unitree, 2 builds: Train a walking policy for Unitree Go2, H1 or G1 in sim · Write your own controller for a Unitree robot with SDK2

Parts notes for this device are in the Hermes Workshop catalog.

What to build next

Builds that share a board, robot or agent with this one.