Mujoco sim of G1 balancing
Week 4 of G1: Trained and deployed my first G1 balancing policy! I put all the code on github in case it helps other folk getting started github.com/jloganols... - the policy is quite jank/brittle, but it's a start!
PID Controller SImulator screenshot
Isaac policy success
Mujoco policy success
Week 3 of G1: Trained a simple policy in both Isaac and Mujoco Playground ('lift one arm'). Took the booties for a test drive. Vibecoded an interactive toy to help me grok PD control (Gemini canvas is great).
3D printed TPU booties for the G1
23DOF rubber-handed G1 in a new Mujoco Playground environment
23DOF rubber-handed G1 in a new Isaac Lab environment
Week 2 with the @UnitreeRobotics G1: Hacked together a 23DOF version of the digital twin and stood it up in Isaac Lab and Mujoco Playground. Prototyped TPU booties so it won't track in dirt after taking walks outside.
Anthropic's latest papers on LLM interpretability are rad (and very accessible!) transformer-circuits...
G1 sdk<>mujoco working - logged the changes you have to make to Unitree's example code here logansrobot.bearblog...
First week with the @UnitreeRobotics G1 - standing up the python SDK was pretty straightforward as was tapping into the LIDAR. Love that it comes folded up in travel a case -- pulling out its lifeless body felt straight out of Weekend at Bernie's
Proud of these two 3D printed doodads I designed last weekend - a lift for our kids' rear-wheel so they can practice pedaling in place and this wall-mount for my mic's boom arm.
