Stringman hangs an open source gripper from four ceiling lines and reaches a whole room

Most robots that can pick something up off the floor pay for every extra centimetre of reach with another motor. Stringman, from Neufangled Robotics, reaches an entire room with four. Its gripper hangs from four lines anchored near the ceiling corners, and the motors reel those lines in and out to place the hand anywhere inside the space they enclose, including over a floor covered in toys, laundry and cables that would stop a wheeled or legged robot.

The company's argument is blunt: "In robotics, price is basically actuator count." Fewer actuators means a cheaper machine, and because all four motors share the lifting, payload grows with their combined torque rather than with the strength of a single joint. The whole design is open source and certified by OSHWA, the Open Source Hardware Association: mechanical CAD, circuit boards and control software are all published.

The trade-off is just as plain. While Stringman is installed, there are lines across your room. Whether that is acceptable depends on your ceiling and on how much you prefer a cluttered ceiling to a cluttered floor.

Four motors instead of a chain of joints

A conventional arm is a chain. The shoulder motor has to hold up everything beyond it: the elbow, the wrist, the gripper and whatever the gripper holds, at the end of a lever. Doubling the reach roughly halves what that joint can lift, which is why desktop arms stay small and why a room-sized arm would be enormously expensive.

Stringman is a cable-driven parallel robot. Each of the four motors sits in an anchor near a ceiling corner and controls the length of one line. All four lines meet at the gripper, so no single motor carries the whole load. Neufangled's comparison uses the same actuator, the DM-H6215, a 1 N·m direct-drive motor, in both designs: in a 0.75 m serial arm that motor sets the limit alone; in Stringman, four of them pull together.

Hanging from above has a second benefit that matters more in a house than in a lab. The robot never touches the floor, so the clutter it is meant to clear cannot block it. The anchors and eyelets define a four-sided work area, and any object inside it can be moved from point to point.

What it can actually lift

The number that governs everything is 28 newtons. That is the tension limit on each line, and Neufangled quotes payload net of the 638 g gripper and its marker, in a reference room of 7 × 7 m with a 3 m ceiling and the anchors in the ceiling corners. Inside that room the usable payload steps through four bands: at least 0.5 kg, 1 kg, 1.5 kg and 2 kg, with the highest values near the middle.

The shape follows from the geometry. Directly under the centre, all four lines pull upward at similar angles and share the weight evenly. Near a wall, the gripper hangs mostly from the two nearest lines, which run at steep angles, so those two reach the 28 N ceiling first. The site has a calculator, under the label "Explore your own room", that redraws the map for other room sizes, and it is worth running before you buy: a smaller or lower room changes the line angles and therefore the numbers.

The 500× claim, and how to read it

Neufangled's headline comparison is about work-area volume per dollar. Measured against a 0.75 m serial arm built from the same DM-H6215 actuators at rated torque, it says Stringman delivers roughly 500 times the work-area volume per dollar at a comparable 2 kg payload. The company also says it reaches the full room with fewer actuators than even the simplest mobile robots.

Two things about that baseline are worth reading closely. The arm figure ignores the weight of the arm's own links, which flatters the arm, since in a real arm the links eat into what the motors can lift. That makes the comparison conservative on that point. On the other hand, volume per dollar measures reach, not speed or placement accuracy. A cable robot swinging a gripper on 3 m of line is a different instrument from a rigid arm bolted to a bench, so check speed and positioning precision against your task before treating the two as interchangeable.

Neufangled pitches the machine as a courier for a home lab rather than a replacement for an arm. Its own illustration shows Stringman ferrying parts between workcells, including a desktop SO-101 arm from TheRobotStudio's open hardware project, so the arms can do fine work while the overhead robot moves things between them.

Installing it: a drill, a ladder and five minutes of calibration

The hardware list is short and domestic:

• A room where the anchors can be fixed to studs or columns.
• A computer to run the controller, with some NPU or GPU; Neufangled says Macs work well.
• A handheld drill.
• A step ladder tall enough to reach the ceiling.

Fitting follows three steps, with picture and video guides on the site.

1. Fix the cleats

   Screw the cleats into wall studs near the ceiling with wood screws, or fasten them to columns with ratchet straps.

2. Hang the anchors and connect the lines

   Each anchor hangs on its cleat; the four lines run from the anchors to the gripper.

3. Run calibration

   A hands-off routine of about five minutes measures the room and adapts Stringman to it.

The stud requirement is the real constraint. Each anchor has to resist the line tension at an angle, so a drywall plug will not do, and anyone renting should check what fixings are allowed before ordering.

Sharing a room with the lines

Neufangled's answer to "will there be lines across my room?" is a straight yes, for as long as the robot is installed. When parked, the gripper and lines sit up near the ceiling. The anchors detach from their mounts without tools, leaving a small white cleat on the wall, so the robot can be taken down without unscrewing anything.

Ceiling fans are fine as long as the anchors are mounted at or below the level of the blades, which makes it physically impossible for the lines or the marker tag to reach them. The original prototype was built in a room with a ceiling fan and still runs there. Low lights off to one side can be handled by shrinking the work area to exclude them.

Safety rests on the same 28 N limit that caps the payload. If a line hits that tension, about the weight of a backpack, all four motors relax and the lines can be pulled loose. Because the lines spend most of their time overhead, Neufangled says collisions with people are uncommon.

Teaching it where things go

Tidying, as Neufangled puts it, is a physical process, not a coding task, and the interface reflects that. You mark destinations, not objects: clip-on fiducial tags go on the laundry bin, the toy box and the bin, and Stringman locates them visually. Individual pieces of clutter need no tags, and anything you never want picked up can be marked off-limits. The same tags double as source and destination markers for ferry jobs between workcells.

The networks trained to control Stringman have learned that the word "toybox" refers to the tag, not to the look of a toy box. That means the tag can turn any container into a toy box, even one that resembles nothing in the training set.

Default models handle automatic grasping and target and destination selection. For new behaviours, Stringman supports control through LeRobot, the open robot-learning framework, with policies including Pi.05, JEPA and Multitask DiT. Training data is recorded from the UI, and for some models, DiT among them, as few as 50 demonstration episodes can be enough to learn a behaviour. Neufangled publishes example recordings from the gripper camera, such as a bedroom session under the dataset name naavox/bedroom-oct15.

Local by default, remote if you ask

Neufangled is explicit that owners should control their own robots. Out of the box, Stringman runs air-gapped on your local Wi-Fi: the UI is served locally, AI models run on your own hardware, and nothing leaves the network. For automation, a Python API and an MCP server, the Model Context Protocol interface that AI agents use to call tools, are aimed squarely at agent-driven workflows.

Remote control is opt-in. Binding the robot through neufangled.com/playroom lets you, or anyone you share access with, drive it from anywhere, and access can be revoked at any time. The company calls the result "a security camera with a hand", says it does not store the information forwarded through its relay, and lets you return to LAN-only mode whenever you like.

Limits and what to watch next

The physics that makes Stringman cheap also sets its ceiling. With every line capped at 28 N, payload tops out around 2 kg in the best part of a generous 7 × 7 × 3 m room and falls toward the walls. That suits toys, laundry and lab parts; it rules out anything heavy, and smaller rooms will change the map, so run the room calculator first.

Then there is the room itself. You need studs or columns near the ceiling, a centre free of hanging fixtures, and tolerance for lines across the space. You also need a computer with an NPU or GPU to run the models locally, which is part of the real cost of ownership even though it buys the privacy.

The open questions are about performance in the field. The 500× figure compares reach and payload per dollar, not speed or precision. The 50-episode figure applies to certain models such as DiT, not every behaviour. And the remote mode's privacy rests on Neufangled's statement that it does not store forwarded data, a promise worth weighing even though the open source code and LAN-only default mean nobody has to take it. How well the trained policies generalise across messy, real homes is the thing to watch as more owners publish their own datasets.

Questions

This article draws on Open Source Alternatives to Reflector? and answers to it (1, 2), on the Stack Exchange network, licensed under CC BY-SA 4.0.