Installation

This project uses four environments:

1. WorldPilot for policy training
2. Cosmos-Policy for Cosmos serving and cache precompute
3. Libero as the LIBERO baseline environment
4. Libero-Plus for public evaluation

The setup below is aligned with this repo's public scripts. Before following the local commands here, prefer checking the upstream installation docs first:

• ABot-Manipulation
• Cosmos-Policy
• LIBERO
• LIBERO-plus

This repo does not use the Docker-first setup described in the official cosmos-policy README. Its public scripts assume local mamba environments and invoke Cosmos with commands such as mamba run -n cosmos-policy python -m cosmos_bridge.cosmos_server.

System packages

Libero-Plus requires a few system-level dependencies. On Ubuntu:

sudo apt-get update
sudo apt-get install -y \
  build-essential cmake git git-lfs curl wget ffmpeg \
  libgl1 libegl1-mesa-dev libgl1-mesa-dri libglib2.0-0 \
  libexpat1 libfontconfig1-dev libpython3-stdlib libmagickwand-dev

1. WorldPilot

Official references:

• ABot-Manipulation

The official ABot-Manipulation setup uses Python 3.10. This repo keeps the same base.

mamba create -n WorldPilot python=3.10 -y
mamba activate WorldPilot

cd /path/to/WorldPilot
pip install torch torchvision --index-url <your-pytorch-cuda-wheel-index>
pip install -r requirements.txt
pip install -e .

VGGT is required by the current Libero config:

pip install -e /path/to/vggt

If your Qwen3-VL stack supports it, install FlashAttention as in the official ABot-Manipulation setup:

pip install flash-attn --no-build-isolation

2. Cosmos-Policy

Official references:

• Cosmos-Policy

The official cosmos-policy repo documents a Docker-first setup around uv sync. The public training and evaluation flow in this repo does not use that Docker path. Use a local mamba environment instead, because cosmos_bridge/run_precompute.sh and the evaluation scripts expect commands like mamba run -n cosmos-policy python ....

mamba create -n cosmos-policy python=3.10 -y
mamba activate cosmos-policy

git clone https://github.com/NVlabs/cosmos-policy.git
cd /path/to/cosmos-policy
pip install -e ".[cu128]"
pip install -r cosmos_policy/experiments/robot/libero/libero_requirements.txt
pip install websockets msgpack

3. Libero

Official reference: LIBERO

The official LIBERO repo uses Python 3.8.13 with the torch==1.11.0+cu113 stack.

mamba create -n libero python=3.8.13 -y
mamba activate libero

pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 \
  --extra-index-url https://download.pytorch.org/whl/cu113

git clone https://github.com/Lifelong-Robot-Learning/LIBERO.git
cd LIBERO
pip install -r requirements.txt
pip install -e .

This environment is not the main public entrypoint of this repo, but it is still useful if you need to compare against the official LIBERO tooling directly.

4. Libero-Plus

Official reference: LIBERO-plus

Libero-Plus is the evaluation environment used by the public scripts in this repo.

mamba create -n libero-plus python=3.8 -y
mamba activate libero-plus

git clone https://github.com/sylvestf/LIBERO-plus.git
cd /path/to/LIBERO-plus
pip install -r requirements.txt
pip install -r extra_requirements.txt
pip install -e .

cd /path/to/WorldPilot
pip install -r examples/LIBERO-plus/eval_files/libero_plus_requirements.txt

Create the local Libero-Plus config:

mkdir -p ~/.libero-plus
cat > ~/.libero-plus/config.yaml <<'EOF'
assets: /path/to/LIBERO-plus/libero/libero/assets
bddl_files: /path/to/LIBERO-plus/libero/libero/bddl_files
benchmark_root: /path/to/LIBERO-plus/libero/libero
datasets: /path/to/libero_plus_datasets
init_states: /path/to/LIBERO-plus/libero/libero/init_files
EOF

Recommended runtime variables for evaluation:

export LIBERO_CONFIG_PATH=~/.libero-plus
export MUJOCO_GL=egl
export PYOPENGL_PLATFORM=egl

Notes

• cosmos_bridge/run_precompute.sh and the Libero-Plus eval scripts assume your Cosmos checkpoint, statistics file, and T5 embedding pickle already exist.
• The public scripts in this repo are written so that you edit the top config block of each .sh file instead of exporting a long list of shell variables.