roboparty_train

English | 中文

Overview

roboparty_train is the Roboparty training workspace for RPO locomotion policies. It keeps Roboparty task definitions, motion-retargeting tools, MuJoCo Sim2Sim scripts, and RSL-RL training entry points outside the upstream Isaac Lab tree.

The repository is organized as a thin workspace with two Git submodules:

• robolab: Roboparty Isaac Lab extension, environments, scripts, and robot assets.
• rsl_rl: Roboparty-compatible RSL-RL dependency snapshot.

Maintainer: RoboParty Support: GitHub Issues

Features

• RPO training environments built on Isaac Lab.
• RSL-RL training, evaluation, and policy export entry points.
• AMP, BeyondMimic, and Parkour workflows.
• MuJoCo Sim2Sim scripts for policy transfer checks.
• Motion retargeting utilities for datasets prepared with GMR.

Requirements

• Python 3.11.
• Isaac Sim 5.1.0 and Isaac Lab 2.3.2.
• Ubuntu 22.04 x64 or Windows x64.

Install Isaac Lab by following the official installation guide. Run the commands below in the Python environment used by Isaac Lab.

Clone

Clone this repository outside the Isaac Lab source tree. Use --recursive so that robolab and rsl_rl are populated immediately:

git clone --recursive https://github.com/Roboparty/roboparty_train.git
cd roboparty_train

If you already cloned the repository and robolab or rsl_rl is empty, initialize the submodules manually:

cd roboparty_train
git submodule update --init --recursive

Install

pip install -e ./robolab
pip install -e ./rsl_rl

Verify that the extension is installed by listing the available environments:

python robolab/scripts/tools/list_envs.py

Usage

Train

python robolab/scripts/rsl_rl/train.py --task=<ENV_NAME> --headless --logger=tensorboard --num_envs=8192

Play

python robolab/scripts/rsl_rl/play.py --task=<ENV_NAME> --num_envs=1

Play AMP

python robolab/scripts/rsl_rl/play_amp.py --task=RPO-AMP-Play --num_envs=1

Play BeyondMimic

python robolab/scripts/rsl_rl/play_bm.py --task=RPO-BeyondMimic --num_envs=1

Play Parkour

python robolab/scripts/rsl_rl/play_parkour.py --task=RPO-Parkour-Play --num_envs=1

To export an ONNX model, set num_envs=1 and add --exportonnx:

python robolab/scripts/rsl_rl/play_parkour.py --task=RPO-Parkour-Play --num_envs=1 --exportonnx

Sim2Sim

python robolab/scripts/mujoco/sim2sim_rpo.py --load_model "{exported/policy.pt model full path here}"

Prepare Motion Data

To prepare datasets for AMP and BeyondMimic, see GMR.

The joint order in datasets produced by GMR follows the robot URDF/XML order, which differs from the order used by Isaac Lab. Prepare a joint mapping file such as robolab/scripts/tools/retarget/config/rpo.yaml, then reorder the joint sequence before training:

python robolab/scripts/tools/retarget/dataset_retarget.py

Troubleshooting

• Empty robolab or rsl_rl directory: run git submodule update --init --recursive.
• Missing Isaac Lab imports: activate the Python environment used by Isaac Lab before installing or running scripts.
• Missing RPO task name: run python robolab/scripts/tools/list_envs.py and copy the exact task id.

References and Thanks

This project builds on the following open-source projects:

• IsaacLab
• rsl_rl
• legged_gym
• legged_lab
• robot_lab
• InstinctLab