Go2 Kinematic Gait & Skating Motion Generator (MuJoCo + MimicKit)

This repository provides a kinematic Walking and Skating gait motion generator for the Unitree Go2 and Go2W robot using MuJoCo, inverse kinematics, and procedural gait controllers. The generated animated motions are exported in MimicKit-compatible format (.pkl) for downstream reinforcement learning using DeepMimic and other methods for getting a policy that generate similar but dynamically feasible motion.

---

Features

• Walking, skating, and static motion modes
• Multiple skating styles (front/back/side/diagonal/two-leg)
• Real-time MuJoCo visualization
• Optional live foot trajectory plotting
• YAML-based configuration with CLI overrides
• Exports loopable MimicKit motion files
• Pure kinematic IK (no physics simulation)

---

1. Create environment

conda create -n mujoco python=3.10
conda activate mujoco 

conda install -c conda-forge mujoco glfw pyyaml numpy matplotlib
pip install pickle5

2. Running the scipt

Running the Script

From the scripts/ directory:

cd scripts

Run with default configuration:

python sim_go2_mimickit.py

3. Supported Modes

Walking

Standard quadruped walking gait.

python sim_go2_mimickit.py --mode walking

Skating

Passive-wheel skating with selectable styles.

python sim_go2_mimickit.py --mode skating --skating_style back_alt

Static

Feet remain fixed (debug/visualization).

python sim_go2_mimickit.py --mode static

4. Additional args

Live Plotting

Enable live plotting of foot trajectories:

--plot true

Plots:

Foot Z-height

Foot X-position

Recording MimicKit Motions

Enable recording:

--record true --record_cycles 4

Saved output:

data/go2_<skating_style>_<mode>_gait_mimickit.pkl

4. Output Format (MimicKit)

Each frame contains:

[root_position (3),
 root_expmap (3),
 joint_angles (...)]

Saved as:

{
  "fps": 200,
  "loop_mode": 1,
  "frames": np.ndarray [T, D]
}