参考链接：
IsaacLab安装教程
miniconda安装教程
DeepRoboticsLab
fan-ziqiGithub仓库
第一期 | 从头训练测试部署一个强化学习策略
第二期 | 训练代码解释，奖励函数设计，仿真环境配置
第三期 | 部署代码解释，程序结构，仿真测试
[具身智能开发第一期] [运控]ROS2结构介绍，以山猫M20盲走运控为例
[具身智能开发第二期] | [Lite3] [Perception] [Hardware Extension] lite3算力传感器扩展

安装Miniconda

miniconda安装教程

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
# Verify your install
conda list
conda --version

安装isaac sim

# Preparing a Python Environment
conda create -n env_isaaclab python=3.11
conda activate env_isaaclab
pip install --upgrade pip

# Installing dependencies
pip install "isaacsim[all,extscache]==5.1.0" --extra-index-url https://pypi.nvidia.com
pip install -U torch==2.7.0 torchvision==0.22.0 --index-url https://download.pytorch.org/whl/cu128

# Verifying the Isaac Sim installation
isaacsim
# 或者
isaacsim isaacsim.exp.full.kit

安装isaac Lab

IsaacLab安装教程

# Cloning Isaac Lab
mkdir isaaclab && cd isaaclab
git clone https://github.com/isaac-sim/IsaacLab.git
# for help
cd IsaacLab/
./isaaclab.sh --help

# install
sudo apt install cmake build-essential
./isaaclab.sh --install # or "./isaaclab.sh -i"
# 以上脚本默认安装所有learning frameworks. 包括rl_games, rsl_rl, sb3, skrl, robomimic.
# 如果只想安装特定框架，可制定参数，如
./isaaclab.sh --install rsl_rl  # or "./isaaclab.sh -i rsl_rl"

# Verifying the Isaac Lab installation
# Option 1: Using the isaaclab.sh executable
# note: this works for both the bundled python and the virtual environment
./isaaclab.sh -p scripts/tutorials/00_sim/create_empty.py
# Option 2: Using python in your virtual environment
python scripts/tutorials/00_sim/create_empty.py

# Train a robot!
# The quickest way to use Isaac Lab is through the predefined workflows using one of 
# our Batteries-included robot tasks. Execute the following command to quickly train 
# an ant to walk! We recommend adding --headless for faster training.
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Ant-v0 --headless
# … Or a robot dog!
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Velocity-Rough-Anymal-C-v0 --headless

# Play a robot!
# run script for playing with 32 environments
./isaaclab.sh -p scripts/reinforcement_learning/rl_games/play.py --task Isaac-Ant-v0 --num_envs 32 --checkpoint /PATH/TO/model.pth
# run script for playing a pre-trained checkpoint with 32 environments
./isaaclab.sh -p scripts/reinforcement_learning/rl_games/play.py --task Isaac-Ant-v0 --num_envs 32 --use_pretrained_checkpoint
# run script for recording video of a trained agent (requires installing `ffmpeg`)
./isaaclab.sh -p scripts/reinforcement_learning/rl_games/play.py --task Isaac-Ant-v0 --headless --video --video_length 200

使用云深处模型训练和play

rl_training

安装DeepRoboticsLab代码

conda activate env_isaaclab
cd ~/isaaclab/
git clone --recurse-submodules https://github.com/DeepRoboticsLab/rl_training.git
cd rl_traning/
python -m pip install -e source/rl_training
# 验证安装，应该能看到4种环境
python scripts/tools/list_envs.py

开始训练train

conda activate env_isaaclab
# Deeprobotics Lite3:
python scripts/reinforcement_learning/rsl_rl/train.py --task=Rough-Deeprobotics-Lite3-v0 --headless
# Deeprobotics M20:
python scripts/reinforcement_learning/rsl_rl/train.py --task=Rough-Deeprobotics-M20-v0 --headless

查看训练过程和结果

可以使用tensorboard --logdir=logs指令查看训练过程和结果，在网页上可以看到各种曲线。

检查训练结果play

注意：
如果希望使用键盘进行控制，需要在以下指令的后面加上--keyboard
若希望在训练中检查训练结果，也可以使用check_point先play查看，在指令最后添加--checkpoint=path_to_model.pt

conda activate env_isaaclab
# Deeprobotics Lite3:
python scripts/reinforcement_learning/rsl_rl/play.py --task=Rough-Deeprobotics-Lite3-v0 --num_envs=10
# Deeprobotics M20:
python scripts/reinforcement_learning/rsl_rl/play.py --task=Rough-Deeprobotics-M20-v0 --num_envs=10

sim to sim仿真验证

Lite3_rl_deploy

Lite3_rl_deploy

注意：
我使用nvidia580的显卡驱动，无法顺利使用mujoco，将版本降低到535和570后可以使用。

配置

conda activate env_isaaclab
# segmentation debug 工具安装
sudo apt-get install libdw-dev
wget https://raw.githubusercontent.com/bombela/backward-cpp/master/backward.hpp
sudo mv backward.hpp /usr/include

# 依赖安装 (python3.10)
pip install pybullet "numpy < 2.0" mujoco
git clone --recurse-submodule https://github.com/DeepRoboticsLab/Lite3_rl_deploy.git

# 编译
cd Lite3_rl_deploy/
mkdir build && cd build
cmake .. -DBUILD_PLATFORM=x86 -DBUILD_SIM=ON -DSEND_REMOTE=OFF
# 指令解释
# -DBUILD_PLATFORM：电脑平台，Ubuntu为x86，机器狗运动主机为arm
# -DBUILD_SIM：是否使用仿真器，如果在实机上部署设为OFF 
make -j

运行
将训练结果policy.onnx（参考路径rl_training/logs/rsl_rl/deeprobotics_lite3_rough/2026-01-12_16-08-42/exported）复制到Lite3_rl_deploy/policy/ppo路径下

# 运行 (打开两个终端)
# 终端1 (pybullet)
cd interface/robot/simulation
python3 pybullet_simulation.py

# 终端1 (mujoco)
cd interface/robot/simulation
python3 mujoco_simulation.py

# 终端2 
cd build
./rl_deploy

操控(终端2)
tips： 可以将仿真器窗口设为始终位于最上层，方便可视化

• z： 机器狗站立进入默认状态
• c： 机器狗站立进入rl控制状态
• wasd：前后左右
• qe：顺逆时针旋转

• pybullet_simulation
  

• mujoco_simulation
  

M20

sdk_deploy
编译

pip install "numpy < 2.0" mujoco
git clone https://github.com/DeepRoboticsLab/sdk_deploy.git

# Compile
cd sdk_deploy
source /opt/ros/humble/setup.bash
colcon build --packages-up-to m20_sdk_deploy --cmake-args -DBUILD_PLATFORM=x86

注意：这里编译的时候可能会遇到不少报错，显示安装包确实，需要根据提示进行安装，python 版本降到3.10

运行

# Run (Open 2 terminals)
# Terminal 1
export ROS_DOMAIN_ID=1
source install/setup.bash
python3 src/M20_sdk_deploy/interface/robot/simulation/mujoco_simulation_ros2.py

# Terminal 2 
export ROS_DOMAIN_ID=1
source install/setup.bash
ros2 run m20_sdk_deploy rl_deploy

使用宇树科技模型训练和play

unitree_rl_lab

安装

下载 RL代码

conda activate env_isaaclab
git clone https://github.com/unitreerobotics/unitree_rl_lab.git
./unitree_rl_lab.sh -i

下载宇树模型文件

# 方式1：使用USD模型文件
git clone https://huggingface.co/datasets/unitreerobotics/unitree_model
# 在source/unitree_rl_lab/unitree_rl_lab/assets/robots/unitree.py文件中修改UNITREE_MODEL_DIR
UNITREE_MODEL_DIR = "</home/user/projects/unitree_usd>"
# 方式2：使用URDF Files [Recommended] Only for Isaacsim >= 5.0
git clone https://github.com/unitreerobotics/unitree_ros.git
# 在source/unitree_rl_lab/unitree_rl_lab/assets/robots/unitree.py文件中修改UNITREE_ROS_DIR
UNITREE_ROS_DIR = "</home/user/projects/unitree_ros/unitree_ros>"

验证环境是否安装完成

# 查看可进行的任务
./unitree_rl_lab.sh -l # This is a faster version than isaaclab

# 开始train任务
./unitree_rl_lab.sh -t --task Unitree-G1-29dof-Velocity # support for autocomplete task-name
# same as
python scripts/rsl_rl/train.py --headless --task Unitree-G1-29dof-Velocity

# 开始play任务
./unitree_rl_lab.sh -p --task Unitree-G1-29dof-Velocity # support for autocomplete task-name
# same as
python scripts/rsl_rl/play.py --task Unitree-G1-29dof-Velocity

Deploy

设置

# Install dependencies
sudo apt install -y libyaml-cpp-dev libboost-all-dev libeigen3-dev libspdlog-dev libfmt-dev
# Install unitree_sdk2
git clone git@github.com:unitreerobotics/unitree_sdk2.git
cd unitree_sdk2
mkdir build && cd build
cmake .. -DBUILD_EXAMPLES=OFF # Install on the /usr/local directory
sudo make install
# Compile the robot_controller
cd unitree_rl_lab/deploy/robots/g1_29dof # or other robots
mkdir build && cd build
cmake .. && make

sim2sim

安装unitree_mujoco

• Set the robot at /simulate/config.yaml to g1
• Set domain_id to 0
• Set enable_elastic_hand to 1
• Set use_joystck to 1.

# start simulation
cd unitree_mujoco/simulate/build
./unitree_mujoco
# ./unitree_mujoco -i 0 -n eth0 -r g1 -s scene_29dof.xml # alternative

cd unitree_rl_lab/deploy/robots/g1_29dof/build
./g1_ctrl
# 1. press [L2 + Up] to set the robot to stand up
# 2. Click the mujoco window, and then press 8 to make the robot feet touch the ground.
# 3. Press [R1 + X] to run the policy.
# 4. Click the mujoco window, and then press 9 to disable the elastic band.

可能是因为我没有手柄，因此在打开unitree_mujoco的时候就会报错

sim2real

./g1_ctrl --network eth0 # eth0 is the network interface name.