YOU CAN BUILD THE 17,950-DOLLAR ROBOT YOU DRIVE WITH YOUR EYES, FOR 632 DOLLARS
Drive a mobile robot with a gripper arm using only your eyes, for 3.5% the cost of the commercial baseline.
by James Dominic Go, Neal Garnett Ong, Carlo Rafanan, Brian Tan, Timothy Scott Chu
RoboticsOpen-hardware
Built withRaspberry Pi3D printing
- difficulty
- ●●●●●
- time
- several weekends
- license
- CC BY 4.0
- repo
- repo FINISHED0 stars
●●●●● · several weekends · CC BY 4.0 · 0 stars · repo FINISHED
WHAT YOU’LL NEED
Jump to section
COMPAREE VERDICT
MoMa is a published hardware research project that shows you can build an eye-controlled mobile manipulator for 632 dollars in parts; the paper names the 17,950-dollar Stretch robot and the 845-dollar Gazepoint GP3 eye tracker as the closest commercial analogues. A mobile base drives forward, back, and turns in place; a two-axis cartesian arm with claw gripper extends and retracts; and a webcam watching your face lets you command the robot by looking at one of seven screen zones for one second. The paper reports nearly 94% accuracy for the four calibrated users, with most individual readings above 96%, and about 85% for three users without calibration. The one failure mode they call out: a user with drooping eyelids that hid the iris when looking down, which made the gaze model struggle. That honesty is rare and useful. The hard parts: laser-cutting acrylic, tuning the gaze calibration per person, and understanding that this is a proof-of-concept, not a mobility aid someone can rely on daily. The DC motors, Raspberry Pi 4 control, and webcam vision are straightforward if you have done robotics before. If you have not, the integration between vision, software, and motion will consume most of your weekends. Most likely to go wrong: assuming the default gaze model will work without per-user calibration, or expecting the mechanical precision of a commercial robot from laser-cut acrylic and 3D-printed plates.
IN THE REPO
GOOD TO KNOW
- —Files hosted on Mendeley Data, not a traditional code repository
- —CAD for laser-cut acrylic plates and 3D-printed parts included
- —Python control code and gaze-tracking model present
- —Bill of materials lists every part with supplier and cost
- —Licence is CC BY 4.0, permits commercial use with attribution
- —Research prototype, not a certified medical or mobility device
Parts to buy
9 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.
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 HardwareX article for context on design decisions and known limitations (Open-access paper with figures, bill of materials, and test results)
- 2.Download the dataset from Mendeley Data for CAD files, code, and BOM(Files hosted on Mendeley Data (doi 10.17632/k7yfn6wdv7.2), linked from the article.)
- 3.Source acrylic sheet and arrange laser-cutting or CNC routing(Plates are the structural frame; precision matters for motor alignment)
- 4.Order the motors, Raspberry Pi 4, lead screws and webcam from the BOM(Bill of materials in the article lists every part with supplier and cost)
- 5.Print motor mounts and gripper parts(STL files in the dataset)
- 6.Assemble mechanics, wire the Raspberry Pi 4, test each axis independently before integration(De-risk by confirming motion works before adding gaze control)
- 7.Set up Python gaze-tracking environment and calibrate per user(Per-user calibration is required; default model will not generalise)
KNOWN ISSUES
- Assuming the gaze model works out-of-the-box: it requires per-user calibration, and the paper shows it fails for users whose eyelids obscure the iris
- Underestimating mechanical precision: acrylic flexes and 3D-printed parts creep; the arm will not be as rigid as the commercial baseline
- Skipping independent axis testing: if you wire everything at once and it does not work, you will not know whether the problem is mechanical, electrical, or software
- Treating this as a daily-use mobility aid: it is a research prototype, not a certified assistive device, and the paper does not claim reliability for unsupervised use
- Ignoring the one-second dwell time: faster commands require lower dwell, which increases false positives; the trade-off is tunable but not automatic
- Buying a low-resolution webcam: gaze tracking needs to resolve the iris clearly; a laptop webcam from 2015 will not cut it
Is this a medical device I can give to someone who needs it?
No. The authors call it a research prototype and do not claim it meets any medical or assistive-device standard. It demonstrates feasibility at low cost, not reliability for daily use.
Do I need the exact motors in the BOM?
No, but you need motors with similar torque and speed, and you will have to adjust mounting brackets if the shaft or body diameter changes. Cheaper motors often have looser tolerances, which shows up as arm wobble.
What controls the robot?
It already does: MoMa uses a Raspberry Pi 4 as its on-board controller, driving the wheel motors and steppers over GPIO and receiving commands over Wi-Fi from the host computer that runs the gaze model.
How long does calibration take per person?
The paper does not give a time. The training script takes 50 images of your eye for each of the seven zones plus 50 of you blinking, 400 images in total, and retrains the model on your eyes.
What if I do not have a laser cutter?
Use an online cutting service or a local makerspace. The paper's bill of materials already counts the two acrylic plates as a fabricated, laser-cut item (about 2,700 Philippine pesos for both), so a service order is part of the 632-dollar total rather than an extra.
Why is the commercial robot so much more expensive?
Stretch is a commercial product, while the 632-dollar MoMa figure is parts cost only: no development time, no warranty, no support, and none of the mechanical refinement the authors do not claim for their acrylic and PLA prototype.
Community builds
No community builds yet — be the first, we feature the best ones.
Discussion1
FROM THE COMPAREE TEAM
The paper reports nearly 94% accuracy for the four calibrated users and about 85% for three users without calibration — and it openly shows the participant whose drooping eyelids tripped the model up on the lower zones. If you built this, what would you test first to know if it will work for your face?
James Dominic Go, Neal Garnett Ong, Carlo Rafanan, Brian Tan, Timothy Scott Chu
Research team at De La Salle University Manila. Published in HardwareX to show that assistive robotics does not have to cost five figures, and documented the failure cases rather than hiding them.
Keep building — related projects
Same category or platform, picked by how close they are to this build.
FULL CADYou can build an open-source underwater research robot for 850 dollars
RoboticsFor expert builders
NO CAD FILESYou can build a palm-sized robot dog that does backflips
RoboticsFor intermediate builders
FULL CADNASA published the plans for its Mars rover so you can build one
RoboticsFor advanced builders
FULL CADHe built a running humanoid robot in 120 days — and gave away every blueprint for free
RoboticsFor expert builders
FULL CADYou can 3D print this open-source robot dog with cycloidal gearboxes
RoboticsFor expert builders
FULL CADYou can 3D print a cube that balances itself on a single corner
RoboticsFor advanced builders
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.
CompareeTEAM22d agoedited
Practical notes from our verification: MoMa is not hosted on GitHub — the design, construction and software files live on Mendeley Data under CC BY 4.0, linked from the HardwareX article, and the robot runs on a Raspberry Pi 4 that takes commands over Wi-Fi from a host computer with a Logitech C920 webcam. The 632 dollar figure is the parts cost from the published bill of materials; it does not cover tools, a laser cutter for the acrylic plates or mistakes. The paper is unusually honest about its limits: for one participant the model struggled because his eyelids drooped low when looking down, hiding the iris from the camera. There is no step-by-step build video — the article has the build instructions and figures, and a testing video is in the Mendeley dataset. The biggest time sink will be per-user calibration with the included training script (50 images of your eye for each of the seven zones, plus blinking) and checking that the 1 second dwell time does not trigger commands you did not mean. 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.