Get the arms ready faster.
Run guided automatic calibration for single or bimanual robots, save named calibrations, and reuse them across skills.
[ OPEN SOURCE · APACHE 2.0 ]
Train, share, and run
robot skills from your browser.
MakerMods Lab turns the SO-101 workflow into a skills workspace. Record your own dataset or start from someone else's, train locally or in the cloud, then run the policy on your robot.
SO-101
single + bimanual
local-first
Hugging Face ready
setup + calibraterecord a datasettrain ACTrun the policy
[ CURRENT FEATURES ]
The tools that keep a robot-learning loop moving.
These are the practical things already in MakerMods Lab today: less setup friction, clearer data, and more ways to build on work that already exists.
See what you actually recorded.
Open a dataset, replay synchronized camera views, and inspect the joint-position chart against the same playhead.
Combine data in the UI.
Merge compatible LeRobot datasets without dropping into a script, then keep the combined result in your library.
Train where it makes sense.
Run training on your own machine or choose cloud compute, with live loss and learning-rate charts in the same workspace.
Do not start every skill from zero.
Import community datasets and policies from Hugging Face, then fine-tune a policy on your own demonstrations.
Put the skill on the arm.
Choose a checkpoint, map the cameras, run the policy, and move straight back into collecting or training when it needs more work.
[ SKILLS · 03 ]
Use other people's work. Make it yours.
Bring in a shared dataset or policy, run it on a compatible setup, or fine-tune it with your own demonstrations. MakerMods Lab is growing into the place where robot skills are found, improved, and passed on.
MAKERMODS SUPPORTED
Sorting socks
Available now · ACT policy
COMING SOON
Opening bottle caps
In training at MakerMods Lab
COMING SOON
Folding towels
In training at MakerMods Lab
COMING SOON
Stacking cubes
In training at MakerMods Lab
[ COMING UP ]
The loop is getting more connected.
These are the next workflows we are building toward. They are not part of the current release yet.
DAgger-style data collection
Run a policy, step in when it drifts, and save those corrections as better training data for the next round.
Remote inference
Keep the robot where it is while policy compute runs elsewhere, with the session controlled through MakerMods Lab.
Remote OpenBooth control
Teleoperate and manage an OpenBooth from somewhere else, so data collection does not require standing beside the rig.
[ QUICK START ]
One command.
Then open the lab.
MakerMods Lab requires Python 3.12 or newer and uv. Install directly from GitHub, connect your SO-101, and launch the local interface.
$ uv tool install "git+https://github.com/makermods-robotics/makermodslab"
$ makermodslab
# UI + API on localhost:8000Python ≥ 3.12SO-101 leader/followerLeRobot datasets