G1 rough-terrain training app (MuJoCo Playground + Brax PPO)

#1
by arminfg - opened
Iterate org

Adds app.py, requirements.txt and README config.

Trains G1JoystickRoughTerrain with Brax PPO on the Space GPU. Training auto-starts on boot in a background thread; the UI is a log tail with start/stop.

After merging, set in Space settings:

  • Sleep time -> never (idle is measured on HTTP traffic, not GPU load)
  • Secret HF_TOKEN (write scope) and variable HF_REPO, so checkpoints are pushed to the Hub -- there is no persistent storage attached.
arminfg changed pull request status to merged

Sign up or log in to comment