B grade Authoritative secondary, traced to origin

A USD 110 arm: why SO-101 is the one to build first

At the low end this is the most literal use of "open source": you can buy the parts, change the structure and replace what breaks. It is also the shortest entry point into the robot learning chain that currently exists.

Form factor: Desktop 5 DOF plus gripper
Structure: 3D-printed parts plus standard hardware; ordinary desktop printer
BOM: From about USD 110 (SO-100 price point; SO-101 a little higher)
Servos: Feetech STS series — bus communication with position feedback
Software pairing: LeRobot (collect, train, deploy in one chain)
Repo: github.com/huggingface/lerobot

Files you actually get: Printed structure plus standard hardware and servos — the buy-the-list-and-start category. The real value is not the files themselves but that a large number of projects use it as the reference point: LeRobot documents its dimensions, servo model and calibration flow, so almost any problem you hit has been hit before and written down.
BOM and delivered cost: The kit lands from about USD 110, servos being most of it. What that price actually buys is the ability to own two — one to run, one to take apart. Half the value of a cheap arm is here: it makes breaking your first one affordable, and breaking your first one is the fastest way to learn.
Printing and machining: Roughly a kilogram of filament; a desktop printer handles the main parts in one go, no industrial equipment needed. Tolerance is where these printed arms go wrong: the same STL off two different machines can come out needing holes reamed or pins sanded. Print two small parts first and check the fit before committing to the whole set.
The hard step: Not assembly — getting calibration and teleoperation stable. The arm goes together quickly; what follows is camera calibration, servo zeroing and consistency in how you record data. Those decide whether the policy you train later is usable at all. This is where most people stall, and most of them blame the arm.
What we checked: We checked: its documented role as the entry-level hardware in LeRobot, the mapping between servo model and bus type, and the price band. Not checked: how a given batch of printed parts actually fits, and which file formats the current release publishes — both move with versions, so treat the repository as the source of truth at the moment you fetch it.

This site's call

If you build one thing, build this. At USD 110, iteration becomes a thing you can afford, and the software chain around it is the most complete one available today. Its ceiling is low — do not expect fine manipulation — but as an entry point to the whole chain, nothing at this price competes.

Why cheap is a capability here, not a compromise

Expensive equipment carries an implicit instruction: do not break it. Hands-on skill grows out of breaking things — disassembling a servo, burning a driver, reaming a hole again teaches more than ten tutorials. What price decides here is how many times you get to practise that.

A sane order of operations

Get the arm moving (teleoperation) — then record a batch of data — then train one small policy — and only then attempt anything requiring precision. Skipping a step in the middle turns into "it just will not move" at the end.

Sources

LeRobot official documentation: entry-level hardware and assembly guidance
Open hardware project release page (structural parts, BOM, servo model)

Objects in this entry

Last checked 2026-09-28