Stringman: A $795 Ceiling-Hung Robot Arm Made of Four Strings and a Gripper

Forum topic · QianXun · 2026-11-04

Summary

Stringman, an open-source robot by Hacker News user nhnifong, hangs a two-finger gripper from four motorized strings anchored to adjacent walls, priced at $795 for the complete unit and certified by OSHWA (us002788). Each of the two active spool boxes holds two DM-H6215 direct-drive motors; the other two walls use passive eyelets, and the four lines converge above the room into a pod with a gripper hanging about 50 cm below. Motion is achieved by varying string lengths, with software-based swing damping, 1080p anchor cameras, and onboard Raspberry Pi hardware. Load capacity is 2.2 kg near the floor, with a 28 N tension cap: exceeding it makes all motors release with damped motion, dropping the gripper to protect the anchors. The system claims ~1000 hours of real-home testing, local Wi-Fi operation, full open CAD/PCB/firmware, MuJoCo simulation, a Hugging Face dataset, LeRobot imitation-learning compatibility (~50 demos per official claim), and an MCP server for agent control. All performance figures come from the official page and have not been independently verified; licensing statements conflict across sources.

Stringman: A $795 Ceiling-Hung Robot Made of Four Strings and a Gripper

Stringman is a robot that flips the usual economics of robotics on its head: instead of paying for arm actuators, it hangs a two-finger gripper from four strings suspended across a room. The complete unit costs $795, is fully open source, and carries OSHWA certification us002788. The author calls the mechanics "tension is all you need."

How it works

Two adjacent walls each mount an active white spool box with two DM-H6215 direct-drive motors (rated ~1 N·m each); the other two walls only have passive eyelets. It is *not* four motors at ceiling corners.
The four strings converge above the room into a marker box, from which a gripper hangs about 50 cm below (638 g pod, two-finger gripper + rotating wrist).
Position is determined by differential string lengths. Load is shared across four strings: 2.2 kg near the floor, 1.9 kg average across the workspace.
Swing is damped by software, not a sixth motor, using visual feedback from two 1080p anchor cameras and one gripper camera.
In a 7×7 m room with 3 m ceilings, the workspace is roughly 147 cubic meters — while a fixed robot arm's workspace is usually measured in liters.

Safety

Tension is the only brake. When any string hits the 28 N limit, the official page states all motors relax and the lines can be pulled loose; the author added on HN that the motors release with a damped motion. The gripper drops rather than the anchors being pulled out of the wall. A standalone wireless emergency-stop button is included.

Software stack

CAD, PCB, and firmware are all open source (GitHub: cranebot3-firmware, author nhnifong). Runs on local Wi-Fi and can be fully air-gapped; model inference uses your own GPU machine (Pi Zero 2W on the gripper, Pi 3A+ at the anchors).
Compatible with LeRobot imitation-learning recipes (Pi.05, JEPA, multi-task DiT). Official claim: ~50 demonstrations suffice to train a pick-and-place policy. MuJoCo simulation is in the repo; datasets are on Hugging Face.
An MCP server ships alongside the Python API, letting Claude or Codex directly command the robot.

Verification notes

The product, author, and OSHWA certification number are all real, but HN scores don't match across snapshots: first-party pages and Algolia both show 824 points / 245 comments, while a third-party snapshot shows 417 points; the widely-cited "360 likes" could not be confirmed. The HN headline rounded the price to $800; the official site says $795.
The $795 figure is the official full-unit price, not a raw bill of materials. Load, tension, room size, and the ~50-demo figure are all official-page claims with no independent reproduction.
The "~1000 hours of testing in real homes" claim appears only as a single summary line on the official site; no hour-by-hour logs or failure-rate statistics are public. The dataset page returned a 429 rate-limit at time of access.
Licensing statements conflict: the OSHWA certification page lists Other in all three fields while mentioning Apache-2.0; secondary sources cite CERN-OHL-W-2.0 and PCB rights reserved. No unified first-party license statement was found.

Community reception

Reactions on Hacker News were split. Some loved the "tension is all you need" framing; others noted the obvious weakness — strings hanging over a living space are a different story if pulled down. Movement speed is 0.4 m/s, and fine manipulation (screwing, plugging) is not in scope: it is a flying porter, nothing more.

The value of the $795 price tag isn't performance. While humanoid robots search for home use cases in the tens of thousands of dollars, Stringman uses the dumbest possible structure to make "something in the room that can move things" dirt cheap, then lets the open imitation-learning ecosystem fill in the brains. The cheaper the structure, the more the intelligence matters — this machine does the math honestly.

---

Sources and caveats: neufangled.com official page and GitHub (cranebot3-firmware, author nhnifong); OSHWA certification us002788. Load, tension, room dimensions, and 50-demo figures are official-page claims, not independently tested. HN post dated Nov 3; safety concerns from comment discussion. LeRobot compatibility and demo counts are official statements, not paper-level claims.