Build challenge · Researched 11 Nov 2026 · Prices USD, date-stamped

Carry 50 lb on four legs
for under $300?

Short answer: no — and it isn't close. The torque gap is roughly 10×. The cheapest motors that fit the budget stall at 1–3.4 N·m per joint; the job needs 9–17 N·m continuous even with short, near-vertical legs, and 25–35+ N·m with normal bent legs. Below is the full math, the three honest build stacks, and what $300 actually buys: a walker that carries 5–10 lb on flat floor.

VERDICT: IMPOSSIBLE AS SPEC'D
5–10 lbclosest honest payload under $300
10–20 lbsalvaged hack, crawl speed — unverified extrapolation
$1,500–$2,500min. realistic budget for a true 50-lb walker (DIY, unproven)

01The torque lab — try to make the math work

Torque is force × lever arm: T = m·g·r. The lever arm r is the horizontal distance from the joint to the foot — not leg length. A perfectly vertical leg needs almost no holding torque; the moment the foot lands out from the hip, the bill arrives. Drag the sliders and watch the requirement run away from every cheap actuator.

Optimistic light build ≈ 22 lb (10 kg). Realistic metal frame + 8 motors + battery: 33–44 lb (15–20 kg).
5 cm = short, near-vertical “crawling table” stance. 15–20 cm = normal bent-knee dog stance.
Walking is sized on 2-leg support (trot). 3-leg crawl is the best case.
Standard practice: ×1.5–2.0 for acceleration, slopes, and foot strike. Hobby servos should only be loaded to ~¼–⅕ of stall continuously.
Total weight—
Static torque / joint—
Peak torque needed—

Worked example, short leg: 172 N per leg × 0.05 m = 8.6 N·m  ·  Same load, foot 20 cm out: 172 N × 0.20 m = 34.4 N·m — 4× the torque for 4× the offset.  ·  1 N·m = 10.2 kg·cm

Cross-checks from real builds: the BOLT quadruped needed ~16 N·m at the hip just to lift itself plus a battery. A University of Minnesota thesis needed a 23.8 N·m commercial actuator for a leg to stand with a 4 kg payload. MIT's Mini Cheetah (~9 kg robot, small payload) uses ~17 N·m peak actuators. Unitree's Go2 runs ~45 N·m joints for an 8 kg payload.

Formula sources: lz-bot.com · RobotShop forum · continuous-load rule: Society of Robots forum

02Actuators — the ladder with a missing middle

A walking quadruped needs at minimum 2 joints per leg — hip + knee — so 8 motors (12 for real turning). Stall torque is not working torque: budget only ~20–25% of stall as continuous. There is no $5–15 actuator that delivers 10+ N·m continuously. The ladder jumps from ~3.5 N·m at ~$16 straight to ~14 N·m at ~$52 — or $10 salvaged, with the driver and feedback costs hidden behind it.

ActuatorStall torqueSpeedCurrentPrice / unit8× costVerdict for 50 lb
MG90S micro servo0.18 N·m0.10 s/60°~0.7 A stall~$1–2~$12Toy only
MG996R servo1.08 N·m (11 kg·cm @6V)0.15–0.19 s/60°2.5 A stall$5–8 (4-pack $29.98)$60–90 (12×)8–16× short even at stall
DS3218MG 20 kg servo2.11 N·m0.14–0.16 s/60°~2–3 A est.$14.99$120–1804–8× short best servo $/torque
DS3235 35 kg servo3.43 N·m0.11–0.13 s/60°3.2 A stall @7.4V$24.49$196–2942.5–5× short 12× alone eats budget
DS5160 60 kg servo5.9 N·m——~$25–35 unverified~$200–280Still short continuous ≈1.5 N·m
Pololu 37D 100:1 gearmotor3.33 N·m100 rpm no-load5.5 A stall$60.95$488Servo torque, 8× price
JGY-370 worm gearmotor, 6 rpm3.57 N·m6 rpm (36°/s)~0.8 A stall$15.87$1272.4× short self-locking, glacial
42GZ495 worm gearmotor, 3 rpm14.2 N·m3 rpm (18°/s)not stated~$52 UK price; US unverified~$416Meets short-leg static 90° swing = 5 s
12 V linear actuator, 150 mm1000 N force12 mm/s~3–5 A est.$25.99 stale price — recheck$208Force yes, walk no
Windshield-wiper motor (used)31.2 N·m measured, one unit~60 rpm no-load45 A stall~$10 junkyard$80Torque yes 1–1.5 kg each, nylon worm wheel strips, no encoder
Hoverboard hub motor (used)~15 N·m peakDirect-drive15 A at peak$10 used / $35 new$80 + driversTorque yes 2 kg each, 6.5″ dia. — unbuildable as a knee

Prices as of 11 Nov 2026. Sources: MG996R spec · DS3218MG · DS3235 · Pololu 37D (verified live) · JGY-370 · 42GZ495 (UK listing) · wiper motor measurement · hoverboard motors

03The three stacks — pick your failure mode

Every build counted the same way — frame, actuators, battery, drivers, controller, fasteners. One fits the budget but not the load. One meets the torque but not the budget — and barely moves. One only works on paper, if the junkyard is kind.

04Every hack, judged honestly

The tricks that get quoted as loopholes. Most of them are real — they just solve a different problem than walking.

HackWhat it savesWhat it breaks
Short legs, feet under hipsTorque drops 3–4× vs long/splayed legs. The single biggest lever — required, not optional.Step height 2–5 cm, belly drags, poor slope stability. Can't clear rubble, cords, or debris. A crawling table, not a dog.
Locking knees / worm gears / stopsHolding current → ~0 standing. Structure carries the static load, so cheap motors can stand with 50 lb.Still needs peak torque to step. Worm efficiency 30–50%, speed 3–6 rpm, vibration can creep a “self-locking” worm, shock goes straight into gear teeth.
Load on central spine, low CoGLess hip-roll torque and tipping; keeps the roll-axis lever arm small.Does nothing for knee/hip pitch torque — the actual bill. A shifting 50-lb load is a tip-over.
CounterweightsCan offset a forward load bias.Adds dead weight the legs must also carry. Net loss. Don't.
Scissor / screw-jack liftCan lift 50 lb vertically, slowly, cheaply.That's a lift, not a walk. Jacks can't swing a leg; one per leg is slower and heavier than worm motors.
Pneumatic legsHigh force per weight from a cylinder.Compressor/valves alone $200+, noisy, leaky, no holding precision. Not credible under $300.
Cable-driven legsMotor mass moves to the body; lower leg inertia. (Cara does this — at $3,300.)Torque requirement unchanged. Rope stretch, routing, and BLDC + FOC driver cost remain.
Used / surplus partsThe only route that puts torque-capable motors inside $300 on paper.No warranty, mismatched units, dead batteries (one salvaged hoverboard pack read ~0 V), LiPo fire risk, no spares. Luck-dependent, unrepeatable.
2-DoF legs, skid-steer turnsSaves 4 motors + drivers (~$60–120) and control complexity.Can't sidestep; turning drags the feet, worse on rough ground. Still needs 8 strong joints.
Six legs insteadTripod gait: 3 feet always down, lower load per leg.The challenge says four legs — and 18 actuators cost more, not less. Ruled out.

Worm self-locking (lead angle < ~5–6°): engineerfix.com · Cara: Hack Club Blueprint

05Frame & gears — your printers help, just not where it hurts

The shop, the FDM printer, and Menards angle genuinely solve the frame. They do not solve the transmission — and the transmission is what fails first.

What fails first, in order

  1. Printed gear teeth — shear at the root; or a hobby servo's internal gears under stall.
  2. Servo shafts & plastic horns — the MG996R's 5.8 mm shaft was never meant for this.
  3. Pivot bolts & printed brackets — single-shear bolts, brackets cracking at bolt holes.
  4. Wiper-motor worm wheels — nylon, strips under shock load.
  5. Battery & drivers — voltage sag and thermal shutdown under stall current.

The motor's rated torque is rarely the first failure. The transmission and mounts are.

Resin gears: no

Standard resin is brittle and fails catastrophically. In a controlled 80:1 planetary test, the resin gearbox died after ~500 cycles — broken teeth, powdered resin — while the PLA gearbox showed very little wear. Tough/ABS-like resin is better, still not a load gear. Resin is fine for housings, sensor mounts, covers. (Hackaday test)

FDM gears: prototypes only

Printed gears reach only 25–35% of bulk-material strength; PLA creeps and softens at 55–60 °C, PETG is tougher but wears, nylon/PA-CF is the best printable option. The rule from the sources: printed gears for motion transmission, not power transmission. (innostation3d)

Metal angle: yes, the thick stuff

SteelWorks 1″ aluminium angle at Menards: $10.48 for 1/16″, $18.98 for 1/8″ (one variant lists $9.33 after an 11% mail-in rebate — a rebate is not an instant discount). The 1/16″ is far too thin for hip brackets at this load. Budget 1/8″ aluminium or steel, ~$25–40 with brackets, bolts, and bearings. Steel is ~3× heavier than aluminium — and weight is torque. (Menards listing)

06Power & control — the quiet budget-eaters

Walking loaded draws ~200 W on a conservative mid-estimate (legged robots use 2–5× the energy per metre of a wheeled one). Try the runtime math:

Before Peukert, sag, and age losses — real runtime will be shorter.
Energy on board111 Wh
Runtime~33 min
Battery cost$76

One pack ≈ 15 min at 200 W. Thirty minutes needs 2 packs ($76); an hour needs 4 ($152) — a quarter to half the entire challenge budget, on batteries alone. A 12 V 7 Ah SLA is ~$25 for ~84 Wh but weighs ~2.5 kg, sags badly at 20–45 A stall, and adds its weight to the torque bill. And don't forget the balance charger: $20–30, missing from most cheap BOMs.

Controller

ESP32 DevKit clone, $4–8 (genuine $10–20). The right pick: Wi-Fi/Bluetooth for phone follow-control, plenty of PWM/I²C. A Pi 4/5 ($35–75+) is unnecessary unless doing camera-based following — which this budget can't support anyway.

Drivers

Servo build: PCA9685 16-ch board, 3-pack $13.99, plus a 10–20 A BEC ($8–15). DC/wiper build: one BTS7960 H-bridge per motor, ~$10 each est. — $64–96 for 8 — plus encoders ($16–32). A DC motor without position feedback is not a servo and cannot walk a gait.

Following you

Remote/joystick follow over the ESP32 costs $0 extra. Genuinely autonomous following on a rough site needs a camera or depth sensor plus a Pi/Jetson ($35–200+) — not credible here. Unitree charges $2,800 for a Go2 Pro with follow mode. Budget for remote control.

Battery (verified live): HRB 3S 5000 mAh · PCA9685: Amazon · ESP32 pricing: sensbench.com

07Speed & terrain — the job site is the hardest case

The short-leg stance that makes the torque math survivable is exactly the design that can't step over job-site debris. Mud, rebar, rubble, slopes, and steps defeat every sub-$300 design — loaded or not.

DesignSpeedStep heightSlopeSurfaces
Servo SpotMicro-class, unloaded0.3–0.5 m/s3–5 cm~10°Flat hard floor, short grass; stalls on gravel
Servo stack with 5–10 lb aboard0.1–0.3 m/s, frequent stalls2–3 cm<10°Flat hard floor only
Large-worm (3 rpm) stack, 50 lb0.02–0.05 m/s2–5 cm<10°Flat concrete/plywood only
Wiper-motor stack, 50 lb (if it works)0.1–0.3 m/s flat3–5 cm~10–15°Flat/firm dirt; mud, loose gravel, stairs: no
Hoverboard QDD build (the $1,500+ one)0.5–1.5 m/s possible5–10 cm20–30°Rough ground possible
Reference: Unitree Go2 ($1,600)Up to 3.5–5 m/s16 cm40°Rough terrain, stairs (manufacturer spec)
Reference: Boston Dynamics Spot ($74,500)1.6 m/s30 cm30°Industrial sites, IP54, 90-min runtime

08The real-world ladder — what payload actually costs

No DIY project, kit, or paper was found that walks with a 20–50 lb payload on a sub-$300 — or even sub-$800 — budget. The heaviest verified DIY payload in the research is 4 kg, on a 23.8 N·m actuator. Here's the ladder, cheapest first:

SpotMicro / SpotMicroESP32 (open source)$150–45012× MG996R servos, Pi/ESP32 + PCA9685. The canonical cheap quadruped. Payload: effectively nil.Project page
Psidog (Hack Club, 2025)≤$800 goalCara-like dog using ≤$40 motors + ~$30 knockoff ODrive ESCs. Explicitly a cost-reduction attempt — not a 50-lb carrier.Blueprint
Unitree Go2$1,600–2,80015 kg robot, ~45 N·m joints. The cheapest commercial quadruped that walks with a real payload.Payload 7–8 kg · max 10–12 kg
Cara (Aaed Musa)$3,30031-lb, mostly 3D-printed quadruped, 8 BLDC cable-driven actuators. No 50-lb payload claim.Project page
DIY quasi-direct-drive joints (Kurapov)€150–250 / jointAssembled QDD joint cost, Sep 2026. Eight joints = €1,200–2,000 for actuators alone — before frame, battery, or drivers. This is the $1,500–2,500 path.Breakdown (verified live)
MIT Mini Cheetah kit$7,59912× QDD actuators, ~17 N·m peak — for a ~9 kg robot with small payload. Kit price excludes batteries.Kit listing
Boston Dynamics Spot$74,500The famous job-site dog. Carries less than this challenge asks: 14 kg.Payload 14 kg
Unitree B2~$100,000~60 kg machine, 360 N·m joints. The commercial proof of what 50-lb-class walking actually takes.Payload 40 kg+ walking
So what should the $300 actually build? Stack A below — a SpotMicro-class walker at ~$236 all-in that genuinely walks and carries 5–10 lb on flat floor: a couple of hand tools, a parts bin, a drink run. The 50-lb carrier is a real project too, but it starts at $1,500–2,500 with QDD actuators, and even that is unproven at that payload. The frame, battery, and controller were never the expensive parts — actuators and their drivers are ~70% of a real build.

Open questions — flagged, not hidden