Yutong Liang, Quanquan Peng, Ri-Zhao Qiu, Xiaolong Wang
This release has been tested on Ubuntu 22.04.5 LTS with an NVIDIA GeForce RTX 4090, NVIDIA driver 550.144.03, Python 3.11, Isaac Sim 5.1.0, and Isaac Lab commit 42e61645c96bac08135566634785cdc87728d5ab.
Create and activate the conda environment.
conda create -n env_isaaclab python=3.11
conda activate env_isaaclabInstall Isaac Sim and PyTorch.
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/cu128Install Isaac Lab.
cd
git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab
git checkout 42e61645c96bac08135566634785cdc87728d5ab
python -m pip install "setuptools<82" wheel
python -m pip install --no-build-isolation flatdict==4.0.1
./isaaclab.sh --install rsl-rlcd
git clone https://github.com/EmptyBlueBox/ConTrack.git
cd ConTrack
python -m pip install -r requirements.txt
python -m pip install -e source/ConTrackOptional wandb login.
export WANDB_API_KEY=XXXXXGenerate simplified collision meshes.
python assets/simplify_mesh.pyConvert the xArm XHand URDF files to USD.
python scripts/tools/convert_urdf.py \
assets/urdf_simplify_collision/xarm_xhand_left-simplified.urdf \
assets/usd/xarm_xhand_left/xarm_xhand_left.usd \
--merge-joints \
--joint-stiffness 0.0 \
--joint-damping 0.0 \
--joint-target-type none \
--headless \
--fix-base
python scripts/tools/convert_urdf.py \
assets/urdf_simplify_collision/xarm_xhand_right-simplified.urdf \
assets/usd/xarm_xhand_right/xarm_xhand_right.usd \
--merge-joints \
--joint-stiffness 0.0 \
--joint-damping 0.0 \
--joint-target-type none \
--headless \
--fix-baseConvert the RB-Y1A Sharpa URDF file to USD.
python scripts/tools/convert_urdf.py \
assets/urdf/rby1a_sharpa.urdf \
assets/usd/rby1a_sharpa/rby1a_sharpa.usd \
--merge-joints \
--joint-stiffness 0.0 \
--joint-damping 0.0 \
--joint-target-type none \
--headless \
--fix-baseReplace <data/xhand/*.h5> with the actual path to the xArm XHand reference trajectory.
CUDA_VISIBLE_DEVICES=0 python scripts/rsl_rl/train.py \
--task Isaac-Xarm-Xhand-Mimic-Manager-v0 \
--data <data/xhand/*.h5> \
--experiment_name XarmXhand \
--num_envs 8000 \
--save_interval 200 \
--headless \
--logger wandb \
--log_project_name ConTrack \
--wandb_description test \
--video_num_envs 9Replace <data/sharpa/*.h5> with the actual path to the RB-Y1A Sharpa reference trajectory.
CUDA_VISIBLE_DEVICES=0 python scripts/rsl_rl/train.py \
--task Isaac-Rby1a-Sharpa-Mimic-Manager-v0 \
--data <data/sharpa/*.h5> \
--experiment_name Rby1aSharpa \
--num_envs 6000 \
--save_interval 200 \
--headless \
--logger wandb \
--log_project_name ConTrack \
--wandb_description test \
--video_num_envs 9Wandb should log the training video every 200 iterations. If you can't see the video, please check the vulkan configuration:
apt-get update
apt-get install -y libxt6 libglu1-mesa vulkan-tools
vulkaninfo --summaryRepository code is released under the BSD 3 Clause License. Third-party assets, dataset clips, and code adapted from upstream projects retain their own terms.
See THIRD_PARTY_NOTICES.md.
