Bootcamp · 21–23 December 2026 · Astana

Humanoid Robot Learning Bootcamp

Three days on the Unitree G1. Program the real robot, train your own walking policy overnight in Isaac Lab, check it in MuJoCo, then run it on the G1.

3days, 09:00–18:00
9seats: 3 teams of 2–3
1 G1per team, plus an RTX 5080 PC
17.5 hof overnight training per team

What your team leaves with

Scripts that run on a real G1: health check, FSM, walking, arm control and the head camera
A walking policy your team trained itself, with one change you chose
That policy checked in Isaac Lab and MuJoCo — and, if it passes the checks, walking on the real G1
A team presentation of your results, and a certificate

Who it's for

Engineers and students who want to go from programming a humanoid to training one.

You need Python, the Linux shell and basic linear algebra. No ROS or reinforcement learning background is needed.

Program

Robots on Day 1 and on Day 3 afternoon. In between, the GPUs train through the night.

Day 1 · Mon 21 Dec

Unitree G1 foundations

Our six-lecture G1 course in one day: from the first connection to a routine on the real robot.

09:00Welcome, safety briefing, teams and stations
09:30Lecture: G1 architecture and safe startup
10:15Lab: first connection, emergency stop and health check
11:15Lab: MuJoCo — arm control and a trained walking policy
13:30Lab: FSM, walking and arm gestures on the real robot
15:15Lab: sensors — RGB-D camera, YOLO detection, LiDAR
16:15Mini-project: a 1-minute routine that combines walking, an arm gesture and the camera
17:30Debrief and quiz

Day 2 · Tue 22 Dec

Isaac Sim and Isaac Lab, then launch the training

Learn the tools in the morning. By 15:15 your team's own training run is going.

09:00Lecture: what Isaac Sim and Isaac Lab are, and how they differ from MuJoCo
09:45Lab: Isaac Sim hands-on — load the G1, inspect its joints, build a scene
11:00Lecture: reinforcement learning for walking robots
11:45Lab: drive a trained G1 walking policy in Isaac Lab
13:30Code tour: Unitree's G1 walking config in unitree_rl_lab
14:15Make your team's change, then a 10-minute test run
15:15Launch the overnight run, then learn to read its training curves
16:15Rehearse Day 3's deployment steps in MuJoCo
17:30Debrief and overnight checklist

Overnight · 15:15 → 08:45

Your policy trains

Each team's run goes from 15:15 to 08:45 — about 17.5 hours on its own GPU. That is long enough to keep Unitree's default rough-terrain training instead of a flat-ground shortcut.

Day 3 · Wed 23 Dec

Check the result: simulation, then the real G1

Three checks, each harder to pass than the last: Isaac Lab, MuJoCo, then the real robot.

09:00Compare all three teams' training curves
09:30Check 1: test the policy in Isaac Lab and export it
10:45Check 2: run it in MuJoCo against a pass/fail checklist
13:00Safety drill on the gantry
13:30Check 3: passing policies go on the real G1 — gantry, slack harness, then free walking with a spotter
15:45Team presentations
16:45Discussion: where simulation and the real robot differed
17:30Wrap-up and certificates

One change per team

Every team trains the same config with one change, so on Day 3 you can see what each change did.

Team A
Keeps Unitree's default setup
The reference: how does the baseline train and walk?
Team B
Penalizes jerky motion more
Does a smoother policy transfer better to MuJoCo and the robot?
Team C
Pushes the robot harder in simulation
Does a policy trained on harder pushes handle the real floor better?

Safety first

A policy goes on the real G1 only if it passes every MuJoCo check. On the robot it starts on a gantry, an instructor holds the remote at all times, and a spotter walks with every moving robot.

MuJoCo checks

Stands still for 30 s without drifting
Walks forward at 0.5 m/s for 20 s without falling
Turns in place and steps sideways, both ways
Recovers from a push
No joint at its limit, no chatter

Details

Dates: 21–23 December 2026, Monday to Wednesday
Hours: 09:00–18:00, with lunch and two breaks
Where: Astana, in person
Group: Up to 9 students in 3 teams of 2–3
Each team gets: A Unitree G1 EDU and a PC with an RTX 5080, Isaac Sim, Isaac Lab and MuJoCo