IsaacLab	this project	Mujoco
	⟹

This repository provides a lightweight deployment of Unitree_RL_Lab training results in Python with Mujoco, without requiring IsaacSim, Unitree_RL_Lab, or IsaacLab installations.

It includes scripts to batch convert training checkpoints into JIT / ONNX models, enabling you to train on a server and easily visualize results locally in Mujoco.

A sample G1 29-DoF walking policy (checkpoint/policy.pt) is provided for testing, and you can replace it with your own trained policies.

• 本仓库无需依赖 IsaaSim, Unitree_RL_Lab 和 IsaacLab 的安装
• 为 Unitree_RL_Lab 的训练结果，提供 Mujoco Python 版本的轻量化部署
• 为 Unitree_RL_Lab 的训练结果，提供批量转换为 JIT / ONNX 模型的脚本
• 应用场景是，在服务器上得到结果训练，拉到在本地，即可直接通过 Mujoco 查看训练结果
• 提供了基础的 G1 29 自由度行走策略(checkpoint/policy.pt)供你尝试，你也可以将其替换为自己训练的策略

1. 参考 unitree_rl_lab ，训练出 29-DoF Unitree G1 行走策略并导出exported/policy.pt

2. 克隆本仓库：

   git clone https://github.com/RoboCubPilot/g1_deploy_mujoco.git

3. 安装必要环境（如果已安装 Isaac Lab 环境可跳过）：

   conda env create -f environment.yml
   conda activate g1_deploy

4. 在 Mujoco 模拟器中运行 Sim2Sim，默认策略路径为 checkpoint/policy.pt：

   python deploy_mujoco.py --policy YOUR_POLICY_PATH

5. （可选）如需将 JIT 格式策略转换为 ONNX 格式：

   python scripts/convert_jit_to_onnx.py --jit-path YOUR_POLICY_PATH --onnx-path OUTPUT_ONNX_PATH

6. （可选）该脚本用于 将 RSL-RL 的训练 checkpoint 批量转换为可部署的 JIT / ONNX 模型，无需安装 IsaacSim 或 IsaacLab。

   python scripts/batch_processing.py --input_path ORIGINAL_CHECKPOINT_PATH --output_path EXPORTED_PATH

   • RIGINAL_CHECKPOINT_PATH: 原始 checkpoint 的路径，可以是单个文件（如 logs/2025-**/model_**.pt）、目录（如 logs/2025-**），或通配符模式.
   • EXPORTED_PATH: 导出模型保存路径（默认：./exported/），也可以指定为任意自定义目录.

1. Train a policy
   Train the 29-DoF Unitree G1 locomotion policy in unitree_rl_lab and export exported/policy.pt

2. Clone this repository

   git clone https://github.com/RoboCubPilot/g1_deploy_mujoco.git

3. Install environment (skip if Isaac Lab is already installed)

   conda env create -f environment.yml
   conda activate g1_deploy

4. Run deployment
   Launch Sim2Sim in Mujoco with the default policy path checkpoint/policy.pt:

   python deploy_mujoco.py --policy YOUR_POLICY_PATH

5. (Optional) Convert JIT → ONNX

   python scripts/convert_jit_to_onnx.py --jit-path YOUR_POLICY_PATH --onnx-path EXPORTED_PATH

6. (Optional) This script batch converts RSL-RL checkpoints into deployable JIT/ONNX models, without requiring Isaac Sim or Isaac Lab.

   python scripts/batch_processing.py --input_path ORIGINAL_CHECKPOINT_PATH --output_path EXPORTED_PATH

   • RIGINAL_CHECKPOINT_PATH: should point to the original checkpoint(s), e.g. a file (logs/2025-**/model_**.pt), directory(logs/2025-**), or wildcard pattern.
   • EXPORTED_PATH: specifies where the exported models will be saved (default: ./exported/). You can override it to any custom folder.

• 🏃‍♂️ Deploy RL policies to Mujoco in seconds
• 🔄 JIT → ONNX conversion supported
• 🔌 Seamless integration with Unitree RL Lab