G1: Week 4
tl;dr
I trained and deployed my own balance policy!
I dumped it on github in case other folks find it useful.
It's quite jank and brittle, per it just falling over with a light push - but hey it's a start!
Learnings
- How to train and deploy a policy that uses the G1 IMU (it's on the pelvis, btw)
- Mujoco Playground is designed to be a showcase/demo repo, not as a foundation for actual work/research
Potential next steps
- Try out LocoMuJoCo
- Teleoperation with the Quest?
- Build the simplest real2sim2real project I could imagine (I just think the community could use that and now I might have the background/skills to do something about it... stay tuned!)