★ PROJECT HEXAPOD ★ INTERNAL
Alternatives & upgrades.
5 DECISIONS · SEEDED FROM DOCS · RESEARCH PENDING
Decision
MG996R vs DS3218MG
CURRENT: MG996R × 18 @ R107.90
MG996R (current)
~R107.90
Pros
- + R107.90 at PiShop; drops to R104.13 at qty 10+
- + Standard hobby servo; every tutorial assumes it
- + Drop-in replacement; gears and horns are standard
Cons
- − 10 kg·cm stall — marginal for an 18-DOF body with PETG frame
- − Analog control; small dead-band causes slight jitter
DS3218MG (upgrade)
~R220
Pros
- + 20 kg·cm stall — 2× MG996R
- + Digital control; tighter dead-band and firmer hold
- + Drop-in mechanically — same horn, same size
Cons
- − ~2× price — 18 units = +R2 000
- − Pushes the BOM over budget
▸Recommendation
Start with MG996R. Upgrade only if Phase 1 reveals body sag under its own weight.
▲Upgrade trigger
Visible femur sag when all six feet are planted, or the tripod cycle can't lift the body reliably.
Decision
Arduino Uno R3 vs ESP32 dev-kit-v1
CURRENT: ESP32 dev-kit-v1 (owned)
ESP32 dev-kit-v1 (current)
owned / free
Pros
- + 520 KB SRAM — holds leg targets, gait tables, DMP buffer, WiFi stack
- + Dual-core 240 MHz with hardware FPU
- + WiFi + BT Classic + BLE built in (Bluepad32 requires BT Classic)
- + Already owned
Cons
- − 3.3 V logic — doesn't matter for I²C + PWM, but watch ultrasonic echo pins
Arduino Uno R3
owned / free
Pros
- + Already owned; simple IDE
Cons
- − 2 KB SRAM — cannot hold 18 leg targets + gait tables + DMP FIFO + serial buffers
- − No WiFi, no Bluetooth
- − Cannot host Bluepad32 or ESPAsyncWebServer
▸Recommendation
ESP32 only. Keep the Uno as a bench-test board or I²C sniffer.
Decision
Antdroid vs Nachum Twersky vs MakeYourPet
CURRENT: Antdroid (antdroid-hexapod/antdroid-hardware)
Antdroid (current)
owned / free
Pros
- + 18× MG996R explicitly in the BOM
- + Parametric CATIA source — dimensions adjustable without redrawing
- + Strong wiki docs; Arduino Mega + Pi reference design
Cons
- − Requires heat-set inserts at load-bearing mount points
Nachum Twersky (fallback)
owned / free
Pros
- + Simpler mechanical design
- + PDF assembly instructions included
- + 3,758 downloads; actively maintained
Cons
- − Requires M2.5 heat-set inserts (+R120)
MakeYourPet (rejected)
owned / free
Pros
- + Visually appealing
Cons
- − Sized for 35 kg·cm coreless servos (ZOSKAY / Feetech FT3550M)
- − Loading MG996Rs requires redrawing every part in CAD
- − Brain is a closed-source Android app (Chica) — no firmware to fork
▸Recommendation
Antdroid. See `docs/11_PRIOR_ERRORS.md` for why MakeYourPet was rejected despite aesthetic preference.
Decision
BUCK-300W-20A vs XL4016
CURRENT: BUCK-300W-20A (Micro Robotics)
BUCK-300W-20A (current)
~R147.20
Pros
- + Real 20 A dual-MOSFET design
- + Handles 20–22 A worst-case servo stall without browning out
Cons
- − Larger; needs airflow under continuous load
XL4016 module (rejected)
~R90
Pros
- + Cheaper; common on Takealot
Cons
- − Genuinely 8 A peak / 5 A continuous / 200 W max despite label claims
- − Will brown out under tripod cycle, reset the ESP32 mid-step, and eventually smoke
▸Recommendation
BUCK-300W-20A only. Never substitute XL4016.
Decision
AI-Thinker ESP32-CAM vs TTGO T-Camera vs Seeed XIAO ESP32-S3 Sense
CURRENT: AI-Thinker ESP32-CAM
AI-Thinker ESP32-CAM (current)
~R203
Pros
- + R203 at Bobshop; standard variant every tutorial uses
- + Flash CameraWebServer and it just works
- + Runs TFLite Micro person_detection at a few FPS
Cons
- − OV2640 consumes nearly every GPIO; don't try to reuse pins
TTGO T-Camera
~R350
Pros
- + Integrated OLED and PIR
Cons
- − Not a drop-in; camera pinout differs
- − Harder to find in SA
Seeed XIAO ESP32-S3 Sense
~R400
Pros
- + Much smaller footprint
- + S3 has more RAM for larger TFLite models
Cons
- − ESP32-S3 breaks Bluepad32 for the main controller — mixing S3 and classic ESP32 is extra complexity
- − Rewires the camera; adapt wiring accordingly
▸Recommendation
AI-Thinker ESP32-CAM only unless stock forces an alternative.