Installation

Installation#

This project uses uv for dependency and environment management. The installation instructions have been tested on Ubuntu 24.04.

1) Install uv#

Follow the official uv installation instructions found here.

Verify your installation:

uv --version

2) Clone MESA#

git clone https://github.com/pairlab/mesa-env.git mesa
cd mesa

3) Install project dependencies#

From the repository root, sync the environment:

uv sync

If you need to export data in LeRobot format, install the optional LeRobot dependencies:

uv sync --extra lerobot

4) Run setup scripts#

Run the setup helper:

./scripts/setup.sh

It runs three scripts from scripts/setup/, which you can also run individually:

  1. setup_robosuite_macros.py configures robosuite’s image convention, which MESA requires.

  2. download_assets.py downloads the simulation assets (scenes, textures, fixtures and objects, about 13 GB) from Hugging Face into mesa/sim/assets/. If you only need single-arm MESA, ./scripts/setup.sh --no-bimesa skips the objects used only by BiMESA (about 8 GB).

  3. download_init_states.py downloads the fixed evaluation initial states for all MESA and BiMESA suites from Hugging Face into mesa/task_suites/init_states/.

To re-download assets or initial states, run the corresponding script with --overwrite:

uv run python scripts/setup/download_assets.py [--no-bimesa] [--overwrite]
uv run python scripts/setup/download_init_states.py [--suites mesa bimesa] [--overwrite]
Contents