MotrixLab

MotrixLab is a reinforcement learning framework for robotics. This application integration package provides the unified mxlab command, environment and robot visualization, and a single installation entry point for the backend-neutral environment framework, policy deployment runtime, RL control plane, built-in FastSAC provider, and MotrixSim simulation backend.

Installation

Use Python 3.10–3.13:

python -m pip install motrixlab

For a beta release, add --pre. All five component packages are pinned to the same release version. The MotrixSim backend installs motrixsim from public PyPI. PyTorch and its dependencies are installed through the component packages; GPU-specific wheel selection and driver setup remain environment-specific.

Usage and scope

mxlab is the command-line entry point for training, playback, environment and robot visualization, export, and deployment:

mxlab --help
mxlab --version
mxlab train --help
mxlab play --help
mxlab view --help
mxlab export-onnx --help
mxlab export-deploy --help
mxlab deploy --help
mxlab deploy inspect artifact=/path/to/policy.deploy

With the selected environment or robot package installed, inspect an environment with random actions or a robot in its default pose:

mxlab view env=<env-id>
mxlab view robot=<robot-id>

Subcommands preserve their component's parameter syntax, including Hydra key=value overrides and configuration-directory options. export-deploy creates an artifact; deploy runs it. mxlab --version is the sole version operation. Install this package for the CLI; the standalone component packages retain their Python APIs and internal callable command adapters, not independent CLI entry points.

The CLI imports only the selected command; top-level help and version do not load Torch or a simulator. Environment and robot visualization belongs to this application package; its view.yaml configuration is packaged in motrixlab.configs. Training, playback, export, and deployment implementations remain in their component packages. Public Python APIs remain in motrix_env_core, motrix_deploy, motrix_rl, motrix_rl_builtin, and motrix_env_motrixsim; the integration package does not re-export them. The built-in RL provider and MotrixSim backend are discovered via package entry points.

This installation does not include the repository's task environments, robot assets, SKRL/RSLRL providers, or deployment backend plugins. Running a task requires installing or supplying its environment and assets separately. For the full repository task collection and GPU-aware source installation, follow the repository quick start.

See the user documentation for framework and training usage.