YOU CAN BUILD A ROBOT THAT HANGS FROM YOUR CEILING AND PICKS THINGS UP OFF THE FLOOR
A robot that hangs from four thin lines in the corners of your room and picks toys off the floor—open hardware, open software, and the creator is honest that it is still a prototype.
RoboticsAutomation
Built withRaspberry Pi3D printing
- difficulty
- ●●●●●
- time
- a month of evenings
- license
- Apache-2.0
- repo
- repo ACTIVE100 stars
●●●●● · a month of evenings · Apache-2.0 · 100 stars · repo ACTIVE
WHAT YOU’LL NEED
Jump to section
COMPAREE VERDICT
Stringman is a room-scale cable robot: two corner anchors each wind two thin lines, a gripper hangs in the middle from all four, and cameras track printed markers to know where everything is. You drive it from a web UI or record demonstrations to train AI policies that pick up objects on their own. The design and software are open under Apache 2.0, the docs under CC BY-SA 4.0, and the creator publishes the BOM, print files and build guides for free. He also sells a kit for 655 dollars if you would rather not source every part yourself. The most likely thing to go wrong is underestimating calibration and tuning. This is not a weekend build: it is a networked set of Raspberry Pi components coordinating line tensions in real time, and the creator says plainly it is a prototype, not an appliance. If you have built a 3D printer from scratch or assembled a quadruped, the complexity will feel familiar. If this is your first robot, expect a steep learning curve on kinematics, camera calibration and policy training. The payoff is working cable robotics at home scale: a robot that uses the whole ceiling.
IN THE REPO
GOOD TO KNOW
- —STLs, Onshape CAD, KiCad PCBs, BOM, and assembly docs are all available for free download from neufangled.com/docs.
- —The design and the control software are under Apache 2.0 (permits commercial use); the documentation is under CC BY-SA 4.0.
- —The creator also sells a DIY kit with all BOM parts for 655 dollars and an assembled unit for 965 dollars—files remain free.
- —Running AI policies requires a host PC with a GPU or NPU; the robot itself does not include compute for inference.
- —The creator states plainly: 'currently a public prototype... can self-calibrate in a new room and run policies, but is not yet reliable enough to be called a home appliance.'
- —This is a complex mechatronics build—four synchronized winches, vision tracking with printed markers, and policy training if you want autonomy.
Parts to buy
5 itemsFrom our check of the build. Exact quantities and part numbers are in the creator’s BOM.
Can I build this?
Build at your own risk. Projects involve tools, electronics and sometimes mains voltage — follow the creator’s safety notes.
Videos
Better Than a Robot Arm? Why I Built a Crane Robot to clean my house
Over Engineer's build story and demo of the Arpeggio version—shows calibration, policy training, and real picking tasks.
More builds like this
All projectsGallery
Start here
Navigation into the creator’s own docs — we don’t rewrite the guide, we route you to the source.
- 1.Read the full documentation at neufangled.com/docs(Start here—the docs walk through the build, calibration, and policy training steps in order.)
- 2.Decide whether to source parts yourself or buy the 655-dollar kit(The BOM is detailed, but sourcing 100+ line items yourself is a project in its own right.)
- 3.Clone the firmware repository and review the setup instructions (You install the motion controller on your PC with pip (nf_robot[host]); it talks to the anchors and gripper over Wi-Fi. Cloning is only needed for development.)
- 4.Print the gripper, motor mounts, and marker targets(STLs are in the docs—print a test marker first to confirm camera visibility.)
Resources
Documentation, files and community threads for this build — we link straight to the original sources and never rehost the creator’s files.
KNOWN ISSUES
- Using a different line than the BOM specifies — line stretch and diameter affect the line-length estimates the robot relies on, so buy the line listed in the BOM.
- Check your room first: you need four reachable corners, a wall stud behind each anchor corner (anchors mount about 13 cm below the ceiling with two wood screws), outlets near the two anchor corners, and nothing hanging from the ceiling in the work area.
- Skipping camera calibration—the vision system is what makes this work. If the cameras cannot see the markers clearly, the robot will not know where it is.
- Assuming the AI policies will work out of the box—you have to record demonstrations and train policies for each task. This is a learning robotics platform, not an appliance.
- Not having a host PC with a GPU—the robot itself does not run inference. You need a separate machine with CUDA or similar to run the policies.
- Expecting consumer-product reliability—the creator is explicit that this is a prototype. It will require tuning, and it will fail in ways a Roomba does not.
Can I build this without buying the kit?
Yes — the BOM, STLs and CAD are free on the docs site. You can source everything yourself, buy just the custom PCBs and specialty wire (55 dollars), or buy the full hardware kit (655 dollars). Sourcing yourself may save money but means many separate orders.
What kind of host PC do I need for the AI policies?
At least 8 cores and 8 GB of RAM. For local AI inference you need some PyTorch acceleration: the README suggests a Ryzen 7 7840HS mini PC or laptop (with an NPU), a Mac mini, or a gaming desktop.
Can it pick up anything besides toys?
It is built to pick things up off the floor and drop them off elsewhere, such as toys and clutter. The docs do not publish a list of what the gripper can and cannot handle, so expect to test it with your own objects.
How long does calibration take?
The creator puts installation at about an hour plus about 30 minutes of setup and calibration (printing markers aside). The robot runs a full calibration from the maintenance menu so it knows where the anchors and eyelets are; a new room needs a new calibration.
Is this safe to run unsupervised?
No. The creator is clear that this is a prototype. It can pick up toys, but it is not reliable enough to leave running while you are out of the house.
Community builds
No community builds yet — be the first, we feature the best ones.
Discussion1
FROM THE COMPAREE TEAM
Three Raspberry Pis, four cables from the ceiling corners and a gripper that picks things up off the floor — sold as a kit or fully assembled. What would you use a ceiling robot for that a floor robot cannot reach?
Nathaniel Nifong
Nathaniel Nifong runs Neufangled Robotics and the Over Engineer YouTube channel. He built Stringman as a research platform for imitation learning and cable robotics at home scale, and he documents both the engineering and the failures. The Arpeggio version is the current iteration.
DISCLAIMER
- Comparee is not the author of the projects featured here. All rights to each project belong to its creator — every page links to the original source, and we never host creators’ files.
- Information is provided without warranty and may become outdated as projects evolve. Prices are indicative bands only — always check the creator’s parts list for current costs.
- Building and operating any project is at your own responsibility. Protective equipment, safe workshop practice and compliance with local regulations are the builder’s responsibility.
CompareeTEAM14d agoedited
Practical notes from our verification: the firmware repository is the controller software for Stringman, and its README points to the build guides and documentation at neufangled.com/docs and to a store that sells the robot assembled or as a hardware kit. The docs include a print guide, anchor and gripper build guides, STL and CAD files and a bill of materials; the design is published under Apache 2.0 and the documentation under CC BY-SA 4.0. A complete robot is three Raspberry Pis on your Wi-Fi — two anchors that each reel two lines on motorised spools, and a gripper — plus a computer running the controller, which installs from pip and pulls in PyTorch and CUDA, around 5.5 GB. The README asks for at least 8 cores and 8 GB of RAM and some PyTorch acceleration for local inference. Before use in a new room, the robot has to run a full calibration so it knows where the anchors and eyelets are. Remote control through the neufangled.com relay is optional, and the README says no video is saved even if you use it. Correction (4 October 2026): we re-checked this page line by line against the project's own repository, documentation and videos, and fixed errors in earlier versions.