Cheap humanoid shell + neck sensors + signal control

All been done. All very expensive. Shouldn't be.

This page is information, not a product. Nothing here is built yet or for sale. It is a thought about useful things: every piece below already exists. The problem is price, not invention. The information is here for someone to actually have it done, or do it for cost — here or anywhere else. No business plan is offered here, and none is invented below. All prices as found Oct 25, 2026; USD unless marked CAD. Prices change — verify before buying.

This is not different from the rest of the work here. It follows the same thread: prosthetics, affordability, progress — development toward fully functional. Its usefulness runs in ways that can't be listed here, and no closed list of uses is claimed below.

Info

Three threads, one idea:

1. The shelf humanoid. A scaled-down humanoid shell, made of plastic, with an OLED face and a working interface — full function, sitting on a shelf, fun to have around. The shell is the cheap part. The system that makes it work is what costs money. Everyone wants something like this. (Yes, plenty of people have the sci-fi horror movies in their heads about humanoids going wrong. A small plastic toy on your shelf that waves, talks, and does what you tell it is a long way from that — and the fact that you built it and control it is the whole point.)

2. Neck contact pads. Pads on the neck that read the electrical signals of muscles as they move — the same principle prosthetics use. Commercial versions are priced far beyond the people who need them. Stripped down to basics, it wouldn't look nice — but the person would have function, and they wouldn't care how it looked, especially if they couldn't afford the real deal.

3. Sending the signals. Those muscle signals are just numbers. They can be sent — to a machine in the room, to a machine somewhere else, to a thing someone else operates. A person with function can set things or have a remote machine do things. Two people without function could each get some function back by sharing control of machines. The sending is the easy part. The decoding and the machine are the work.

What you need to know first: surface EMG (sEMG) reads muscle electricity through pads stuck on the skin. Nothing goes into the body. It is the same sensing commercial prosthetics use, minus the clinic, the fitting, the certification, and the price tag that comes with all of that.

Costs

The shelf humanoid — what exists now

The cost driver, stated plainly: Poppy's motors are 60% of its $8,000–$9,000. Swap smart servos for hobby servos and the same shape of robot drops under $200 (Aster) — but it loses walking and strength. The shell is plastic and cheap; the system — motors, control, software — is where the money goes.

Parts for a stripped plastic build (sourced prices, then an honest total):

• MG996R hobby servo (metal gear, 13 kg-cm): ~$7 each in a 2026 price guide; $6–$15 range across sellers. 16 of them (Aster count) ≈ $112.
• SG90 micro servo (light joints): ~$2.50 each.
• OLED eye modules (0.96" SSD1306, 128×64): $2.99 each (Robot Pi Shop, Oct 2026); 4-pack CAD $25.70 (RobotShop Canada). Two eyes ≈ $6.
• A full face is cheapest as a recycled phone screen (Aster used a Galaxy S4 as the face) — no new part to buy.
• Controller board (Arduino-class or ESP32-class), filament, power supply, battery: estimates only — roughly $25 for the board, $25–$50 filament, ~$20 power, based on general maker pricing, not a single sourced listing.
• Rough total for an Aster-style shelf build: ~$190–$230 in parts (estimate built from the sourced prices above), plus a 3D printer or a print service. That buys a puppet-like humanoid with a face and a working interface — not a walking one. Walking is what multiplies the price.

Neck sEMG — the sensors

A basics-only DIY neck rig (estimate from the sourced parts above): one MyoWare board ($42.95) + pads ($9.95) + a microcontroller (estimate ~$10–$20) ≈ $65–$75 for one channel. Two to three channels (the count the wheelchair research used) ≈ $120–$180 (estimate). Strap it on with a band and a dab of electrode gel. Ugly. Works.

What the commercial world charges for the same sensing

The gap, in one line: the sensing board is $43. The commercial arm using the same principle is $20,000 and up. The difference is not the sensor — it is fitting, certification, clinic time, motors, liability, and markup. e-NABLE already proved the at-cost distribution model works for body-powered hands; nobody has done the same at scale for EMG ones. That is the opening.

Signal relay — the sending

• The wireless MyoWare kit ($124.95, above) already sends EMG over an ESP32. Wireless EMG transmission is a commodity feature, not a research project.
• Beyond that, sending muscle signals is sending numbers over radio, Wi-Fi, or the internet. There is no exotic link to invent. The cost is in the board at one end and the machine at the other.

Theory

How sEMG works. When the brain tells a muscle to move, motor units fire and the muscle gives off a tiny electrical signal. Pads on the skin pick it up, a board filters and rectifies it, and out comes a voltage that rises with how hard the muscle is working. That voltage becomes a control input: cross a threshold and a servo moves, a grip closes, a command fires. One channel gives you one signal (on/off, or a level). More channels, placed on different muscles, give you more distinct commands — left/right, open/close, up/down.

Why the neck. The neck and shoulder muscles (trapezius, sternocleidomastoid, platysma) stay under voluntary control for many people whose arms and legs do not. That is why researchers keep coming back to the neck: it is the control source that is still there. Neck pads are also easy to wear — a band, not a socket fitted to a limb.

Why a humanoid shell is cheap and a humanoid system is not. Plastic is cheap. Filament by the kilo is cheap. What costs money: motors with enough torque and precision (Dynamixels are ~$200 each; Poppy needs 25), power delivery (a stalled MG996R can spike to 2.5 A — power is a real design problem, not an afterthought), and the software that turns intent into motion. Scale the robot down and the torque problem shrinks with it — a shelf toy does not need to hold up its own body weight on two legs, which is the hardest and most expensive problem in humanoids. That is why the shelf version is honest and the walking version is expensive.

Why relayed control is plausible. An EMG signal, once digitized, is indistinguishable from any other sensor reading. It can drive a local servo, a prosthetic hand, a wheelchair, a robot arm across the room, or a machine across the world — the transport does not care. Two people sharing function means two signal sources mapped onto one machine's commands, or one person's signals driving a machine that does something for another person. The mapping is software. The machine is hardware. Neither step violates physics; both steps need careful, boring engineering (thresholds, false triggers, fatigue — muscles get tired, and tired muscles change their signal).

The perception barrier, briefly, plainly. People have the horror movies in their heads. Fine — that is a story problem, not an engineering problem. A small plastic thing on a shelf with a face you made, that does what you tell it, is a toy and a tool. If it helps someone move something they couldn't move, nobody is thinking about the movies anymore.

Reality

What is proven today (with receipts):

• Open-source humanoids you can build now. InMoov (free files, ~$900–$1,000 upper body in older reports, $1,200–$3,500 in a 2026 guide), Poppy ($8,000–$9,000, 25 smart servos), Aster (under $200, hobby servos, phone face, can't walk). All real, all built by hobbyists.
• Cheap EMG boards you can buy now. MyoWare 2.0 ($42.95), SpikerBox kits ($99.99–$299.99). Snap on pads, read a voltage, drive a servo. This is a weekend project, not a lab.
• Neck EMG control, in research, with real participants. A 3-electrode neck/shoulder sEMG system on an ESP32 drove a powered wheelchair with five commands: 100% accuracy in healthy participants, 96.75% in participants with disabilities (JoVE report). Neck/shoulder sEMG drove a TetraSki adaptive ski chair as well as sip-and-puff control, with less mental workload (peer-reviewed, PMC). Neck EMG moved a computer cursor for a person with high cervical tetraplegia (case study). Single-sensor frequency-domain classification worked on trapezius and platysma signals in tetraplegia (MDPI Sensors, 2022). Wearable sensors decoded hand-movement intention in people with complete cervical spinal cord injury years after injury (PMC, 2024).
• EMG teleoperation of robot arms, in research, for years. Surface EMG drove a robot arm in real time in 3D space (Artemiadis & Kyriakopoulos, 2010). A 4-electrode EMG system teleoperated a robot arm in 2D catching movements (2006 paper). NASA JPL's BioSleeve — dry cloth electrodes on a sleeve — teleoperated mobility, a manipulator arm, and camera pan/tilt, tested indoors and outdoors (US patent US10449672B2). A 2026 teleoperation paper runs sEMG + IMU at 200 Hz with under 20 ms model latency — in a lab.
• At-cost distribution, proven by someone else. e-NABLE: $30–$50 of materials, $0 to the recipient, 8,000+ devices, 100+ countries. The model works. Its hands are body-powered, not EMG — but the distribution proof is the point.
• Commercial humanoids are finally getting cheaper, and still aren't cheap. Unitree G1 at $13,500–$16,000 is the floor for a real walking humanoid, and it is a research platform, not a household helper.

What is NOT proven (honest):

• A buyable "neck pad kit." Nobody sells one. Neck control exists as research prototypes and as DIY boards you stick wherever you want. The neck version has to be assembled from general-purpose parts.
• The two-people-share-function version as a daily product. One person's EMG driving a machine is proven. Two people's signals shared onto one machine, or one person's signals doing something for another person as a product — not demonstrated anywhere found. The pieces are all proven; the combination as a lived, daily thing is not.
• A cheap humanoid that walks. Aster proves cheap and humanoid-shaped. Poppy proves open-source and walking-ish. Nothing proves cheap AND walking AND useful at the same time. The shelf version is the honest target; walking is a different budget.
• Any of this as a certified medical device. None of the DIY prices include fitting, clinical testing, certification, or liability. That is exactly why the commercial versions cost what they cost — and exactly why the DIY versions are for hobby, learning, and at-cost help, not a medical claim. This page is assistive-technology information only. It is not medical advice, not a treatment, and not a recommendation for any individual. Anyone building for a specific person's needs should work with that person's clinicians.

The real useful use. For someone who can't move a limb because of a neural disconnect: sense the signals above the disconnect, send them to stimulate below it, and muscles could develop and function could be restored. This is the direction functional electrical stimulation and neural-bypass research already works in — sensing on one side of an injury, stimulating on the other. It is a research direction, not a promise for any individual, and outcomes vary widely. It is also why the cheap version matters: the people who need this most are the least able to pay for the expensive one.

What would make this real for someone:

1. A neck band with 2–3 MyoWare-class boards and a microcontroller, mapped to simple commands (open/close, left/right, on/off).
2. A shelf humanoid (Aster-style) that those commands drive — arms, head, face, voice — as a visible, fun proof that the signals work.
3. The same signals driving something useful: a gripper, a wheelchair command set, a remote machine.
4. All of it documented at cost, with files, so anyone anywhere can repeat it without asking permission or paying a premium.

What's proven / what's not — honesty box

Sources (read Oct 25, 2026)

• InMoov project: inmoov.fr (free STL files, Arduino + MyRobotLab); cost reports: element14 community ("about $900 USD, minus the cost of a 3D printer"), EDN ("parts only cost about $900"), 2026 hobbyist guide at articsledge.com (upper body $1,200–$3,500; full bipedal $5,000–$15,000; Digit $150,000–$250,000).
• EZ-InMoov Humanoid Robot Hardware Kit: $2,799.99 USD, plastic parts not included — ez-robot.com store.
• Poppy Humanoid: poppy-project.org / GitHub README ($8,000–$9,000, ~60% actuators, 25 Dynamixels, ~7 h assembly); Hackaday (€7,500–€8,000 estimate, ~$200 per MX-28 actuator).
• Aster (Poppy remix, MG996R servos, Arduino Uno, phone face, under $200, can't walk): Hackster, 2026.
• MG996R servo: ~$7 each (electricalflux 2026 price guide); $6–$15 range (blog.kpower.com); stall spike 2.5 A (electricalflux).
• SG90 micro servo: ~$2.50 (electricalflux 2026 guide).
• OLED modules: 0.96" SSD1306 $2.99 (Robot Pi Shop); 4-pack CAD $25.70 (RobotShop Canada); 0.42" module $10.98 USD (RobotShop).
• MyoWare 2.0: $42.95 sensor, $79.95 Basic Kit, $124.95 Wireless Kit, pads $9.95/10-pack (SparkFun store + tutorials, Oct 2026); Development Kit CAD $285.41 (Canadian reseller).
• Backyard Brains: Muscle SpikerBox Classic $189.99, Human SpikerBox $299.99, Spiker:bit $99.99 (backyardbrains.com, Oct 2026).
• Commercial prosthetics: $20,000–$40,000 (3D-printed myoelectric arm thesis); $30,000–$150,000 myoelectric range (2025 market report); Ontario manufacture $20,000–$140,000 and ~£250,000 advanced device (PMC costing study); Atom Touch target ~$25,000, not for sale, no FDA clearance as of 2025.
• e-NABLE: materials $30–$50, $0 to recipients, 8,000+ devices, 100+ countries (3dprintwithpurpose.com; techxlab.org). Devices are body-powered, not EMG.
• Neck sEMG wheelchair: 3-electrode system, trapezius + sternocleidomastoid, ESP32, five commands, 96.75% accuracy (participants with disabilities) — JoVE Visualize report.
• Neck sEMG TetraSki: sEMG vs sip-and-puff, no performance difference, less cognitive load — PMC (IEEE Int Conf Rehabil Robot).
• Neck EMG cursor control, high cervical tetraplegia: case study (The Free Library / journal report); implanted sensors in that study, not surface.
• Single-sensor sEMG classification, trapezius/platysma, tetraplegia: MDPI Sensors 2022, 22, 1939.
• Wearable decoding of hand intention in complete cervical SCI: PMC, 2024 (PMCID PMC11449141).
• EMG teleoperation: Artemiadis & Kyriakopoulos 2010 (real-time 3D robot arm, surface EMG); 2006 2D catching teleoperation paper (horc-lab PDF); BioSleeve wireless EMG sleeve, teleoperation modes (US10449672B2); 2026 sEMG+IMU teleoperation paper, 200 Hz, ≤20 ms latency (Intelligent Robotics).
• Commercial humanoids: Unitree G1 $13,500 base / $16,000 launch (2026 buyer's guide, Notebookcheck); G1 overseas ~$27,300 from $50,000, BOM ~$8,976 (SemiAnalysis via ChinaBizInsider, 2026); R1 Air $4,900; 1X NEO $20,000 early access (NextTechSearch 2026 guide); Western platforms $100,000–$200,000 (iaexpertos, 2026).

Could not verify this pass (stated, not papered over): a current 2026 InMoov total (the $900 figure is from older project reports); any off-the-shelf neck-EMG kit; the two-person shared-function version anywhere; Raspberry Pi / Arduino board prices (used only as labelled estimates); Canadian retail prices and shipping/tax for most items (most figures are USD).