This repository contains sim-2-real pipeline for Unitree Go2.
This project is based on IsaacLab 2.0.2 and IsaacSim 4.5.0.
Inside asig_sim2real folder is code that defines the training environment.
Inside deployment folder is code for deploying to MuJoCo or on real robots.
Inside rsl_rl folder is the rsl_rl library, which implements the PPO algorithm and neural networks. We modify the code based on version 2.3.1.
The most important file is asig_sim2real/tasks/go2_env.py,
where the training environment is defined,
including observations, rewards, curriculums, and so on.
The go2_env_cfg.py file defines some configurations for the training environment.
The go2_ppo_cfg.py defines PPO configurations.
In the motor folder, a motor model with randomized motor strength is implemented.
In the terrain folder, we define the terrain for the training environment.
Inside the deployment folder you can find C++ code for deploying the trained policy.
Run run_build.sh to build and run locomotion_node to deploy a policy.
This is the source code of the rsl_rl library. We use version 2.3.1. We don't need to modify anything here.
If you want to cite this, please consider cite this project, which has both locomotion and obstacle avoidance policies.
Contributor: Qihao Yuan