🤖 G1 Sim2Crawl

Keyboard-Controlled Sim2Sim Crawling for Unitree G1

G1 Sim2Crawl is a MuJoCo-based sim-to-sim deployment toolkit for the Unitree G1 humanoid, enabling interactive crawling control from trained Isaac Lab policies — using only a keyboard or gamepad.

This project makes it possible to:

• Train locomotion in Isaac Lab
• Export policies
• Run them in MuJoCo
• Control the robot live using keyboard or gamepad

No Isaac Sim runtime is required during playback.

---

✨ What Makes This Project Different

Compared to the original G1 Crawl pipeline, this repo adds:

• ⌨️ Keyboard control (no gamepad required)
• 🎮 Full gamepad parity
• 🔁 Live policy switching
• 🛡️ Safety & gain control
• 📦 Plug-and-play MuJoCo sim2sim
• 📚 Bilingual documentation
• 🎥 Reproducible demo pipeline

This turns a research policy into a hands-on, testable, reproducible robotics system.

---

🧬 Project Lineage

This project is derived from and builds upon:

jloganolson/g1_crawl
https://github.com/jloganolson/g1_crawl

lez666-g1-sim2crawl extends the original project with:

• MuJoCo-based sim2sim deployment
• Keyboard-based control (no gamepad required)
• Improved documentation and reproducibility

All credit for the original G1 crawling policy and Isaac Lab training framework belongs to the original authors.

---

🚀 Quick Start (Keyboard Sim2Sim)

Option 1: Automatic Installation (Recommended)

# Clone the repository
git clone <your-repo-url>
cd lez666-g1-sim2crawl

# Run the automatic setup script
./setup.sh

# Activate environment and run
conda activate g1-crawl
python sim2sim_mj/run_sim2sim_keyboard.py

Option 2: Manual Installation

# Create conda environment
conda create -n g1-crawl python=3.10 -y
conda activate g1-crawl

# Install dependencies
pip install -r sim2sim_mj/requirements.txt

# Run simulation
python sim2sim_mj/run_sim2sim_keyboard.py

For detailed installation instructions, see INSTALL.md

No controller required — just use your keyboard.

---

⌨️ Default Key Bindings

Action	Keys
Forward / Backward	↑ / ↓
Strafe Left / Right	← / →
Rotate Left / Right	Z / C
Mode	I (stand) · J (damped) · K (crawl)
Switch Policy	SPACE
Gain Up / Down	Q / A
Print Gains	H
Exit	ESC

---

📘 Keyboard Control Guide

• English → sim2sim_mj/README_keyboard_en.md
• 中文 (Chinese) → sim2sim_mj/README_keyboard_zh.md

---

🎮 Gamepad Mode

You can also use a standard controller:

python sim2sim_mj/run_sim2sim.py

Supports:

• Analog velocity
• Policy cycling
• Mode switching
• Safety exit
• CSV diagnostics

---

🧪 Training (Isaac Lab)

Minimal headless training

python scripts/rsl_rl/train.py --task g1-crawl --headless

🧪 Play a Trained Policy

python scripts/rsl_rl/play.py --task g1-crawl --headless --video --enable_cameras

🤖 Real Robot Deployment

The deployment/ folder contains robot-side execution pipelines with:

• PD control
• Watchdogs
• Hardware interfaces

⚠ Always validate in simulation before deploying on real hardware.

---

📂 Key Paths

Purpose	Path
Training logs	logs/rsl_rl/
Sweep results	sweep-logs/
Sim2Sim policies	sim2sim_mj/policies/
MuJoCo model	sim2sim_mj/scene.xml

---

🛠 3D Printed Parts

• Head: Onshape link
• Arms: Onshape link
• Print files: 3d-printed-parts/