Files

wertyuilife f2fcb966b9 update readme.

2026-04-17 21:58:53 +08:00

4.3 KiB

Raw Blame History

go2_rl_robotlab

Overview

Trains the Unitree Go2 robot using MoE-CTS in IsaacLab and deploys the trained policy to MuJoCo for simulation transfer (Sim2Sim).

It is a reproduction of go2_rl_gym, adapted to the RobotLab / IsaacLab ecosystem.

Installation Guide

1. Install IsaacLab

Install IsaacLab 2.3.0 release by following the installation guide.

After installation, your environment should satisfies:

isaacsim <= 5.1.0.0   # tested on 5.1.0.0
isaaclab <= 0.53.1    # tested on 0.53.1
isaaclab-rl <= 0.4.7  # tested on 0.4.7

Higher version may cause conflicts with our customized rsl_rl==3.3.0 and robot_lab==2.3.0.

2. Install Customized RSL-RL and RobotLab

We uses a customized version of rsl_rl and robot_lab. Install them in editable mode:

python -m pip install -e source/robot_lab
python -m pip install -e source/rsl_rl

3. Install MuJoCo (Optional, for Sim2Sim)

To enable MuJoCo-based simulation:

pip install mujoco  # tested on 3.4.0 and 3.6.0

Training and Evaluation

Run the following commands:

# Train
python scripts/rsl_rl/train.py --task=RobotLab-Go2-v0 --headless

# Evaluate
python scripts/rsl_rl/play.py --task=RobotLab-Go2-v0

Configuration

The training pipeline can be configured at two levels: task-level Python configuration files and runtime arguments passed to the training script.

Task-Level Configuration

The default task settings are defined in the following files:

Environment configuration

source/robot_lab/robot_lab/tasks/go2/env_cfg.py

RL algorithm configuration

source/robot_lab/robot_lab/tasks/go2/rsl_rl_cfg.py

Task registration

source/robot_lab/robot_lab/tasks/go2/__init__.py

Runtime Overrides

In addition to the default configuration files, train.py and play.py supports several command-line arguments for runtime overrides:

python scripts/rsl_rl/train.py \
   --task=RobotLab-Go2-v0 \
   --headless \
   --experiment_name <YOUR_EXP_NAME> \
   --run_name <YOUR_RUN_NAME> \
   --num_envs <NUM_ENVS> \
   --checkpoint <PATH_TO_CHECKPOINT>

For more details, refer to the robot_lab repo.

MuJoCo Sim2Sim

Run the deployment script:

python deploy/deploy_mujoco/deploy_go2.py

Controller Behavior

Automatic detection: If a controller is connected, control mode is enabled automatically.
Fallback mode: If no controller is detected, default commands from the config file are used.

Controller Mapping

Input	Function
`LX / LY`	Forward / lateral velocity
`RX`	Angular velocity (steering)

Switching Simulation Scenarios

Modify xml_path in:

deploy/deploy_mujoco/config/go2.yaml

# Flat terrain
xml_path: "{ROOT_DIR}/resources/go2/flat.xml"

# Stairs
xml_path: "{ROOT_DIR}/resources/go2/stairs.xml"

# Boxes
xml_path: "{ROOT_DIR}/resources/go2/boxes.xml"

# Custom
xml_path: "{ROOT_DIR}/resources/go2/your-custom-scene.xml"

Differences from `go2_rl_gym`

Different terrain composition
Different tracking reward formulation (fixed sigma vs. dynamic sigma)
Lack domain_rand: randomize_motor_zero_offset
Lack domain_rand: randomize_motor_strength

TODO

Try replacing ActionManager-level delay with DelayedPDActuatorCfg

Acknowledgements

This repository would not exist without the following open-source projects:

isaac_lab: Unified framework for robot learning built on NVIDIA Isaac Sim.
rsl_rl: Reinforcement learning algorithms.
robot_lab: RL Extension Library for Robots, Based on IsaacLab.
mujoco: High-performance CPU physics simulator.

Related publications implemented in this repo:

CTS: Concurrent Teacher-Student Reinforcement Learning for Legged Locomotion

4.3 KiB Raw Blame History