Use in Your Own Project

The LeRobot Installation Guide guide clones lerobot_trossen and runs the commands from inside that checkout. That is the quickest way to get going, but if you are building your own application you usually want the Trossen LeRobot plugins as dependencies of your own project instead. This lets you control your own environment, pin your own versions, and decide up front which policy families to install.

This page shows how to depend on the Trossen plugin packages. It also shows how to bake in the LeRobot extras (pi, smolvla, async) so you never need the uv run --with prefix at run time.

The Packages

lerobot_trossen is a uv workspace that publishes two plugin packages:

• lerobot-robot-trossen: the follower-robot plugin (registers the widowxai_follower_robot, bi_widowxai_follower_robot, and mobileai_robot robot types).

• lerobot-teleoperator-trossen: the leader / teleoperator plugin.

Both require Python ≥ 3.12 (they depend on lerobot >= 0.5.1). They are not published to PyPI, so depend on them directly from Git.

Add the Dependencies

1. Add the plugin packages from the lerobot_trossen repository.

   Each package lives in its own subdirectory of the workspace:

   uv add "lerobot-robot-trossen @ git+https://github.com/TrossenRobotics/lerobot_trossen.git#subdirectory=packages/lerobot_robot_trossen"
   uv add "lerobot-teleoperator-trossen @ git+https://github.com/TrossenRobotics/lerobot_trossen.git#subdirectory=packages/lerobot_teleoperator_trossen"
   

   Equivalently, declare them in your pyproject.toml:

   [project]
   requires-python = ">=3.12"
   dependencies = [
       "lerobot-robot-trossen",
       "lerobot-teleoperator-trossen",
   ]
   
   [tool.uv.sources]
   lerobot-robot-trossen = { git = "https://github.com/TrossenRobotics/lerobot_trossen.git", subdirectory = "packages/lerobot_robot_trossen" }
   lerobot-teleoperator-trossen = { git = "https://github.com/TrossenRobotics/lerobot_trossen.git", subdirectory = "packages/lerobot_teleoperator_trossen" }
   

2. Resolve and install:

   uv sync
   

Declaring Policy Extras

The plugin packages pull in a lean LeRobot install. This is enough to teleoperate, record, replay, and run ACT policies, but without the heavier optional stacks. VLA policies and async inference need additional LeRobot extras:

Extra	Needed for
pi	π₀ / π₀.₅ policies (pulls in transformers / peft)
smolvla	SmolVLA policies
async	Async Inference (the grpcio policy-server / robot-client transport)

When you work from a clone of lerobot_trossen, these are layered in per command with uv run --with "lerobot[pi]>=0.5.1". In your own project you can instead declare them once, so they are always installed and the --with prefix is no longer needed.

Add the extras alongside the plugin packages:

uv add "lerobot[pi,smolvla,async]>=0.5.1"

Or in pyproject.toml:

[project]
dependencies = [
    "lerobot-robot-trossen",
    "lerobot-teleoperator-trossen",
    "lerobot[pi,smolvla,async]>=0.5.1",
]

Note

Only declare the extras you actually use; each one adds significant install size. If you only run ACT policies, the base plugin dependencies are enough and you can skip this section entirely.

Running Commands

With the dependencies declared, run the LeRobot entry points directly through your project’s environment, with no --with prefix:

uv run lerobot-record \
    --robot.type=widowxai_follower_robot \
    --robot.ip_address=192.168.1.4 \
    --robot.id=follower \
    --robot.cameras='{
        cam_main: {type: intelrealsense, serial_number_or_name: "0123456789", width: 640, height: 480, fps: 30}
        }' \
    --display_data=false \
    --dataset.repo_id=${HF_USER}/eval_trossen_ai_solo_test \
    --dataset.single_task="Grab the cube" \
    --policy.path=${HF_USER}/my_pi05_policy

See Training and Evaluating a Policy for the full evaluation workflow. See Async Inference for the policy-server / robot-client flow.