Jarvis Humanoid Robot — Complete Build Guide

00

Full system architecture

Jarvis Humanoid — Kinematic tree + intelligence layers

  ┌─────────────────────────────────────────────────────┐
  │            INTELLIGENCE LAYERS (Wi-Fi)              │
  │  [Gemini API free]  [Mobile Phi-3]  [Laptop LLaMA]  │
  └───────────────────────┬─────────────────────────────┘
                          │ WebSocket / HTTP
                          ▼
  ┌─────────────────────────────────────────────────────┐
  │                 HEAD / CAMERA NODE                  │
  │   ESP32-CAM · VC-02 voice AI · 0.96" OLED face       │
  │   2× MG90S neck servos  ← ch 0 (pan) · ch 1 (tilt) │
  └───────────────────────┬─────────────────────────────┘
                          │
  ┌─────────────────────────────────────────────────────┐
  │                     TORSO BLOCK                     │
  │  PCA9685 (16-ch driver) · MPU6050 (gyro) · MB102 (pwr) │
  │  Battery pack · TP4056 charger module               │
  └──────────┬──────────────────────────────┬───────────┘
             │                              │
    ┌────────┴──────┐              ┌────────┴──────┐
    │   LEFT ARM    │              │   RIGHT ARM   │
    │ ch2 Shoulder  │              │ ch5 Shoulder  │
    │ ch3 Elbow     │              │ ch6 Elbow     │
    │ ch4 Gripper   │              │ ch7 Gripper   │
    │ (3× MG90S)    │              │ (3× MG90S)    │
    └───────────────┘              └───────────────┘
                          │
  ┌─────────────────────────────────────────────────────┐
  │                  PELVIS HIP FRAME                   │
  │    ch8 Hip-Roll LEFT (MG996R)                       │
  │    ch9 Hip-Roll RIGHT (MG996R)                      │
  └──────────┬──────────────────────────────┬───────────┘
             │                              │
    ┌────────┴──────┐              ┌────────┴──────┐
    │  LEFT LEG     │              │  RIGHT LEG    │
    │ ch10 Hip Pitch│              │ ch13 Hip Pitch│
    │ (MG996R)      │              │ (MG996R)      │
    │ ch11 Knee     │              │ ch14 Knee     │
    │ (MG90S)       │              │ (MG90S)       │
    │ ch12 Ankle    │              │ ch15 Ankle    │
    │ (MG90S)       │              │ (MG90S)       │
    │               │              │               │
    │ ▓▓▓▓▓▓▓▓▓▓▓▓ │              │ ▓▓▓▓▓▓▓▓▓▓▓▓ │
    │ 10cm×6cm FOOT │              │ 10cm×6cm FOOT │
    │ IR sensor ↓   │              │ IR sensor ↓   │
    └───────────────┘              └───────────────┘

    SAFETY: HC-SR04 ultrasonic (front) · 2× IR cliff (feet)

Brain stack

ESP32-CAM — main MCU + Wi-Fi + camera
VC-02 — offline speech recognition
MPU6050 — 6-axis gyro/accelerometer
PCA9685 — I2C 16-channel servo driver
0.96" OLED — face expressions

Actuators

4× MG996R — high-torque hips (11 kg/cm)
12× MG90S — all other joints (metal gear)
Total: 16 servos, 16 DoF

Power

2× 18650 Li-ion cells (series = 7.4V)
TP4056 USB-C charging module
MB102 → 5V/3.3V logic rail
Motor rail: direct from battery
~2.5 hr normal operation

01

Physical dimensions — every part

📐

Scale reference: Jarvis is approximately 2× OnePlus Nord CE in height (~30–32 cm). All measurements are for cardboard/tin sandwich construction. Adjust ±5mm based on your servo mounting method.

Side-view skeletal dimensions (all in cm)

   ┌──────────────┐  ← 6cm wide × 5cm tall × 4cm deep
   │  HEAD BOX    │     Cardboard box · Camera + OLED inside
   └──────┬───────┘
          │ neck 1.5cm joint
   ┌──────┴───────┐  ← 9cm wide × 8cm tall × 5cm deep
   │    TORSO     │     All electronics inside
   │  (ESP32 etc) │
   └──────┬───────┘
   ╔══════╩═══════╗  ← 12cm wide · hip pivot block
   ║  HIP FRAME   ║     2× MG996R here (Roll joints)
   ╚═══╤══════╤═══╝
       │      │  ← 3.5cm gap between legs
┌─────┴┐    ┌┴─────┐
│THIGH │    │THIGH │  ← 7cm long × 3cm wide each (MG996R pitch)
└──┬───┘    └───┬──┘
   │5cm          │5cm
┌──┴───┐    ┌───┴──┐
│ SHIN │    │ SHIN │  ← 7cm long × 2.5cm wide (MG90S knee)
└──┬───┘    └───┬──┘
   │3cm          │3cm
┌──┴────────────┴──┐
│ FEET (flat base) │  ← 10cm × 6cm each · 5mm thick
│  IR sensor below │
└──────────────────┘

ARMS (each):
  Upper arm: 5cm long × 2cm wide
  Lower arm: 5cm long × 2cm wide
  Gripper claw: 4cm × 3cm (scissor-style cardboard)

TOTAL HEIGHT: ~30–33cm
TOTAL WEIGHT TARGET: 450–550g

Body part	Width	Height/Length	Depth
Head box	6 cm	5 cm	4 cm
Neck bracket	3 cm	1.5 cm	3 cm
Torso block	9 cm	8 cm	5 cm
Hip frame	12 cm	4 cm	5 cm
Thigh (each)	3 cm	7 cm	2.5 cm
Shin (each)	2.5 cm	7 cm	2 cm
Foot plate (each)	10 cm	6 cm	0.5 cm
Upper arm (each)	2 cm	5 cm	2 cm
Lower arm (each)	2 cm	5 cm	2 cm
Gripper (each)	4 cm	3 cm	1.5 cm

Cardboard sandwich rule

Every structural piece = 3–4 layers of cardboard glued together. Single layer will flex and fail under servo torque. Double-sided tape between layers first, then hot glue edges.

Weight budget

Head + neck: ~60g
Torso + electronics: ~120g
Hip frame + 2 MG996R: ~130g
Each leg: ~50g × 2 = 100g
Each arm: ~20g × 2 = 40g
Total target: ≤ 550g

02

All components — full specifications

Component	Model	Qty	Key specs	Purpose	Source	Price
MCU + Camera	ESP32-CAM	1	240MHz dual-core · 4MB flash · OV2640 · 802.11n · I2C/UART/GPIO	Main brain, Wi-Fi, visual tracking	Robu.in	₹600
Voice AI	AI-Thinker VC-02 Kit	1	US516P6 32-bit · 2MB SPI flash · 150+ offline cmds · UART · USB-C · Mic + Speaker	Offline speech recognition — no internet needed	ESC Labs (₹700) or Robu dev kit	₹700
Servo driver	PCA9685	1	I2C · 16 channels · 12-bit PWM · 5–6V · up to 25 MHz	Controls all 16 servos from 2 wires (SDA/SCL)	Robu.in	₹220
Gyro/IMU	MPU6050	1	6-axis (accel + gyro) · I2C · 3.3V/5V · addr 0x68	Balance feedback for PID loop	Local / Robu	₹110
OLED face	SSD1306 0.96"	1	128×64 px · I2C · 3.3V/5V · addr 0x3C	Robot face expressions + status	Robu.in	₹200
Hip servo ×4	MG996R TowerPro	4	11 kg/cm · Metal gear · 55g · 4.8–7.2V · 180° · 55×20×38mm	Hip Roll (×2, pelvis) + Hip Pitch (×2, thigh)	Quartz Components ₹254/pc	₹1,016
Body servo ×12	MG90S (metal gear)	12	2.2 kg/cm · Metal gear · 13.4g · 4.8–6V · 180° · 32×12×29mm	Neck×2, Shoulders×2, Elbows×2, Grippers×2, Knees×2, Ankles×2	Local market / Robu ₹90/pc	₹1,080
Power module	MB102	1	5V / 3.3V dual rail · AMS1117 regulator · breadboard mount	Clean logic voltage for ESP32, PCA9685, sensors	Robu.in	₹70
Battery	18650 Li-ion × 2	2	Samsung/LG/Panasonic · 2600–3000mAh · 3.7V each · 65×18mm	Series = 7.4V, direct motor rail	Local electronics market	₹220
Charger	TP4056 (Type-C)	1	1A charge · overvoltage protect · LED status · micro/USB-C	Charge robot via phone charger	Local / Robu	₹30
Cliff sensor	IR obstacle ×2	2	2–30cm · Digital output · 3.3V/5V · adjustable	Mounted under each foot — stops at table edge	Local market	₹70
Front sensor	HC-SR04	1	2–400cm · ±3mm accuracy · 5V · TRIG/ECHO pins	Front obstacle avoidance	Local market	₹50
Voice sensor	TF-LC02 LiDAR (optional)	0/1	3–200cm · VCSEL · 3.3V · UART/I2C · compact	Optional precision distance for trash-grab task	Robu.in	₹867

⚠️

MG996R buying test: Lift the motor in your hand — genuine 11 kg/cm metal gear should feel heavy (~55–60g). Rotate the shaft manually with 2 fingers — it should resist strongly. If it spins easily, it's a fake. Ask specifically: "MG996R TowerPro, metal gear, 11kg/cm".

03

Cardboard + tin structure guide

Materials needed

Amazon boxes — double-walled (brown thick type)
Tin sheet — from old containers / oil tins (for leg bones — more rigid)
Hot glue gun — structural bonding
FeviQuick / Cyanoacrylate — layer lamination
Heavy utility cutter — metal blade type
Binding wire — backup servo horn mount
Black electrical tape — wire insulation
Isopropyl alcohol — remove glue if servo misaligned

🔴

Critical rule: Calibrate ALL servos to 90° in code BEFORE gluing anything. If you glue a joint at the wrong angle, you'll need IPA to unstick it and redo.

Servo slot cut technique

Step 1: Measure servo body
  MG996R: 40mm × 20mm body
  MG90S:  23mm × 12mm body

Step 2: Cut slot in cardboard
  ┌────────────────────────────┐
  │   CARDBOARD PIECE          │
  │    ┌──────────────┐        │
  │    │  servo slot  │← exact │
  │    │  (snug fit)  │  match │
  │    └──────────────┘        │
  └────────────────────────────┘

Step 3: Test fit first (no glue yet)
  → servo should be snug but removable

Step 4: Apply hot glue ONLY on edges
  → Never glue over servo body center
  → Never let glue touch wire exit

Step 5: Horn connection
  → Cardboard limb presses onto horn
  → Hot glue + binding wire loop
  → Screws through horn holes preferred

NEVER hot glue directly on PCBs or
electronic module surfaces.
Use double-sided foam tape instead.

Part-by-part construction guide

🧠 Head box (6×5×4 cm)

Cut 6 faces from cardboard (3-layer sandwich each)
Front face: cut 2 circles for eyes (18mm diameter)
Left eye: black glass marble / dummy lens
Right eye: ESP32-CAM lens aligned from inside
Bottom face: cut MG90S slot (pitch servo goes here)
OLED display slot at "mouth" area (bottom-center front)
Glue ESP32-CAM inside with foam tape on its PCB back
Leave top open for air circulation

🔗 Neck bracket (L-shape, 3×1.5×3 cm)

Cut one L-shaped bracket from 3-layer cardboard
Top horizontal arm: holds MG90S pitch motor (Up/Down)
Bottom vertical arm: holds MG90S yaw motor (Left/Right)
Pitch servo horn → glues to bottom of head box
Yaw servo body → glues into torso top center slot
Use binding wire through all horn screw holes

📦 Torso (9×8×5 cm)

Largest single piece — 4-layer sandwich
Top center: MG90S yaw servo slot (from neck)
Front panel: reserved for VC-02 speaker grille
Inside: mount ESP32-CAM board (foam tape), PCA9685 (foam tape), MB102
Foam tape > hot glue for PCBs — removable when debugging
Left/right sides: arm shoulder servo slots (facing outward)
Bottom: connects to hip frame with 2 long screws or binding wire

🦴 Hip frame (12×4×5 cm)

Widest structural piece — use tin sheet + cardboard hybrid
Center-left: MG996R Roll slot (left hip)
Center-right: MG996R Roll slot (right hip)
Both MG996Rs face downward, shafts pointing out sides
Battery pack mounts on rear face of this frame (low CoG)
TP4056 USB-C port faces rear for easy charging access

🦵 Leg (thigh + shin, 7cm each)

Thigh: tin strip + cardboard backing (more rigid)
Top of thigh: MG996R Hip Pitch servo slot
Bottom of thigh: MG90S Knee servo slot
Shin: 2-layer cardboard + tin reinforcement
Bottom of shin: MG90S Ankle servo slot
Ankle horn connects to foot plate center

👣 Foot plate (10×6×0.5 cm)

Use 4-layer thick cardboard or thin plywood (preferred)
Center-rear: ankle servo horn attachment point
Underside front: IR cliff sensor (flat, facing down)
Edges: reinforce with electrical tape to prevent fraying
Width (10cm) gives stability — do NOT make narrower
Anti-slip: stick small rubber pads or foam pieces on underside

💪 Arm (shoulder + elbow + gripper)

Upper arm: 5cm strip, MG90S shoulder at top, MG90S elbow at bottom
Lower arm: 5cm strip, connects from elbow horn
Gripper: two cardboard "fingers" hinged at center
Gripper servo pushes one side to create clamping action
Gripper opening: ~4cm max (enough for small objects)
Reinforce arm joints with binding wire loops

🎭 Face plate (front of head)

Cut from black cardboard or paint black
Two equal eye circles: 18mm diameter, symmetrical
Left eye: epoxy a black marble or dummy lens in hole
Right eye: ESP32-CAM lens aligned exactly at hole center
Mouth slot: 26mm × 14mm for 0.96" OLED
Optional: black transparent visor (X-ray film strip) over both eyes

04

Complete wiring diagram

🔴

Power isolation rule: Motors and logic MUST share GND but NOT share +V. Motor rail goes directly from battery → PCA9685 V+. Logic rail goes battery → MB102 → 5V → ESP32-CAM/VC-02/sensors. Mixing these will cause ESP32 brownouts and random resets.

Complete wiring schematic

┌──────────────────────────────────────────────────────────────────────┐
│                      POWER DISTRIBUTION                              │
│                                                                      │
│  [18650 × 2]──+── (+7.4V) ──────────────────── PCA9685 V+           │
│   Series        \                                                    │
│   (7.4V)         └── (+7.4V) ─── MB102 IN ─── 5V OUT ─── ESP32-CAM │
│                                                │                    │
│  [GND bus] ─────────────────────────────────── MB102 GND            │
│      │                                         │                    │
│      └── PCA9685 GND                           └── 3.3V OUT (NC)    │
│      └── ESP32-CAM GND                                              │
│      └── VC-02 GND                                                  │
│      └── MPU6050 GND                                                │
│      └── All sensor GNDs                                            │
│                                                                      │
│  [TP4056] ── USB-C IN ── Charge 18650s                              │
└──────────────────────────────────────────────────────────────────────┘

┌──────────────────────────────────────────────────────────────────────┐
│                    I2C BUS (SDA / SCL)                               │
│                                                                      │
│  ESP32-CAM GPIO21 (SDA) ─────────────────────────────────────────   │
│          │                                                           │
│          ├── PCA9685 SDA (addr 0x40 default)                        │
│          ├── MPU6050 SDA (addr 0x68)                                │
│          └── OLED SSD1306 SDA (addr 0x3C)                           │
│                                                                      │
│  ESP32-CAM GPIO22 (SCL) ─────────────────────────────────────────   │
│          │                                                           │
│          ├── PCA9685 SCL                                            │
│          ├── MPU6050 SCL                                            │
│          └── OLED SSD1306 SCL                                       │
│                                                                      │
│  NOTE: Add 4.7kΩ pull-up resistors on SDA and SCL to 3.3V          │
└──────────────────────────────────────────────────────────────────────┘

┌──────────────────────────────────────────────────────────────────────┐
│                    UART BUS (VC-02 Voice)                            │
│                                                                      │
│  ESP32-CAM GPIO1  (TX) ───── VC-02 RX                               │
│  ESP32-CAM GPIO3  (RX) ───── VC-02 TX                               │
│  ESP32-CAM 5V         ───── VC-02 VCC                               │
│  GND BUS              ───── VC-02 GND                               │
│                                                                      │
│  VC-02 → Speaker (built-in) · Mic (built-in on kit)                 │
└──────────────────────────────────────────────────────────────────────┘

┌──────────────────────────────────────────────────────────────────────┐
│                   GPIO PINS (ESP32-CAM)                              │
│                                                                      │
│  GPIO4  → IR sensor LEFT FOOT (digital in)                          │
│  GPIO13 → IR sensor RIGHT FOOT (digital in)                         │
│  GPIO14 → HC-SR04 TRIGGER                                           │
│  GPIO15 → HC-SR04 ECHO                                              │
│  GPIO0  → Boot mode (leave HIGH during operation)                   │
│                                                                      │
│  NOTE: GPIO16 is not usable on ESP32-CAM (connected to PSRAM)       │
│  NOTE: GPIO12 must be LOW at boot (do not pull HIGH)                │
│  NOTE: Flash LED is on GPIO4 — avoid for IR sensor if using flash   │
└──────────────────────────────────────────────────────────────────────┘

┌──────────────────────────────────────────────────────────────────────┐
│                   SERVO SIGNAL WIRES                                 │
│                                                                      │
│  ALL SERVO GND ────────────── GND BUS                               │
│  ALL SERVO V+  ────────────── PCA9685 V+ (7.4V motor rail)          │
│  ALL SERVO SIG ────────────── PCA9685 CH0 through CH15              │
│                                                                      │
│  Per-servo connector: 3-pin JST/Dupont                              │
│    Brown/Black = GND                                                 │
│    Red         = V+ (servo power)                                   │
│    Orange/Yellow = Signal (PWM from PCA9685)                        │
└──────────────────────────────────────────────────────────────────────┘

Wire-by-wire connection table

PCA9685 Servo Driver connections

PCA9685 VCC→MB102 5V output (for IC logic only)

PCA9685 GND→Common GND bus

PCA9685 SDA→ESP32-CAM GPIO21

PCA9685 SCL→ESP32-CAM GPIO22

PCA9685 V+→Battery (+7.4V) DIRECTLY — this powers all servos

PCA9685 GND (power row)→Common GND bus

MPU6050 Gyroscope connections

MPU6050 VCC→ESP32-CAM 3.3V pin (NOT 5V)

MPU6050 GND→Common GND bus

MPU6050 SDA→ESP32-CAM GPIO21 (shared I2C bus)

MPU6050 SCL→ESP32-CAM GPIO22 (shared I2C bus)

MPU6050 INT→ESP32-CAM GPIO34 (interrupt, optional)

OLED SSD1306 connections (face display)

OLED VCC→ESP32-CAM 3.3V or 5V (both work)

OLED GND→Common GND bus

OLED SDA→ESP32-CAM GPIO21 (shared I2C bus)

OLED SCL→ESP32-CAM GPIO22 (shared I2C bus)

VC-02 Voice Module connections (UART)

VC-02 VCC→MB102 5V output

VC-02 GND→Common GND bus

VC-02 TX→ESP32-CAM GPIO3 (RX) — cross-connect TX→RX

VC-02 RX→ESP32-CAM GPIO1 (TX) — cross-connect RX→TX

HC-SR04 Ultrasonic (front obstacle)

HCSR04 VCC→5V from MB102

HCSR04 GND→Common GND bus

HCSR04 TRIG→ESP32-CAM GPIO14

HCSR04 ECHO→ESP32-CAM GPIO15 (use 1kΩ+2kΩ voltage divider — ECHO is 5V, ESP32 is 3.3V)

IR Cliff Sensors (2× under feet)

IR Left VCC→3.3V or 5V

IR Left OUT→ESP32-CAM GPIO4

IR Right VCC→3.3V or 5V

IR Right OUT→ESP32-CAM GPIO13

Both GND→Common GND bus

✅

Debugging tip: Before connecting all servos, connect just ONE servo to PCA9685 channel 0 and run a sweep test. If it moves correctly, then add servos one by one. This isolates wiring problems early.

05

PCA9685 servo channel mapping

All 16 channels mapped. PCA9685 default I2C address is 0x40. Set PWM frequency to 50Hz for servos.

00

Neck Pan / Yaw

MG90S · Left/Right head turn

01

Neck Tilt / Pitch

MG90S · Up/Down head nod

02

Left Shoulder

MG90S · Arm forward/back

03

Left Elbow

MG90S · Arm bend

04

Left Gripper

MG90S · Claw open/close

05

Right Shoulder

MG90S · Arm forward/back

06

Right Elbow

MG90S · Arm bend

07

Right Gripper

MG90S · Claw open/close

08

Left Hip Roll

MG996R · Lateral balance (pelvis)

09

Right Hip Roll

MG996R · Lateral balance (pelvis)

10

Left Hip Pitch

MG996R · Thigh forward/back swing

11

Left Knee

MG90S · Lower leg bend

12

Left Ankle

MG90S · Foot angle / floor contact

13

Right Hip Pitch

MG996R · Thigh forward/back swing

14

Right Knee

MG90S · Lower leg bend

15

Right Ankle

MG90S · Foot angle / floor contact

ℹ️

PWM values for MG90S/MG996R: 0° = 150 counts (0.5ms pulse), 90° = 375 counts (1.5ms pulse), 180° = 600 counts (2.5ms pulse). Set PCA9685 frequency to exactly 50Hz using: setPWMFreq(50)

06

Power system design

Power flow diagram

[18650 Cell 1] ── series ── [18650 Cell 2]
                    │
                  +7.4V  (fully charged: +8.4V)
                    │
         ┌──────────┼──────────────────────┐
         │          │                      │
         ▼          ▼                      ▼
  [TP4056]    [PCA9685 V+]           [MB102 IN]
  USB-C         Motor Rail              │
  Charging     (powers all            ┌─┴─┐
                 16 servos)          5V   3.3V
                   │                 │
                   │            ┌────┼────┐
                   │           ESP32 VC02 Sensors
                   │           CAM
                   │
            [ALL SERVO V+ wires]

COMMON GND: Battery─ connects to PCA9685 GND,
            MB102 GND, all sensor GNDs, ESP32 GND

Battery life calculation:
  2× 18650 2600mAh in series = 2600mAh total capacity
  Average current draw (normal operation): ~1000mA
  Light use (mostly standing): ~400mA
  Heavy walking: ~1800mA (peak)
  ─────────────────────────────────────────────────
  Normal use: 2600 / 1000 = ~2.6 hours
  Light use:  2600 / 400  = ~6.5 hours  
  Full walk:  2600 / 1800 = ~1.4 hours

18650 battery selection

Brand: Samsung 25R, LG HG2, Panasonic NCR
Capacity: 2600–3000mAh (higher = longer)
Max continuous discharge: 10A+ (high drain type)
Buy from known source only — fakes are common
Get a genuine 18650 holder (with leads) for easy swap

Safety rules

Never short Li-ion cells — fire risk
TP4056 auto-cuts at 4.2V per cell (safe charging)
Never discharge below 3.0V per cell (2 cells = 6.0V cutoff)
If battery gets warm during use — stop and check
Use XT30 connectors (not thin dupont) for battery leads

07

Software stack + code structure

ESP32 firmware (C++)

Arduino IDE + ESP32 board package
Libraries: Adafruit PCA9685, MPU6050, Wire, SSD1306, WebSocket
FreeRTOS for multitasking (Core 0 = net, Core 1 = motors)

Laptop server (Python/Node)

FastAPI (Python) or Node.js + Express
WebSocket server for real-time motor commands
OpenCV for face/object detection
Gemini API for AI responses

Dashboard (Next.js)

Your strength — MERN stack applies here
Live servo sliders for manual testing
Camera feed viewer
Gait sequence editor (dance moves JSON)

ESP32 starter firmware skeleton

                // jarvis_firmware.ino — Phase 1 starter
                #include <Wire.h>
                #include <Adafruit_PWMServoDriver.h>
                #include <Adafruit_MPU6050.h>
                #include <Adafruit_SSD1306.h>
                #include <WiFi.h>
                #include <WebSocketsServer.h>

                // ── Servo PWM constants ──────────────────────────
                #define SERVO_MIN 150 // 0.5ms → 0°
                #define SERVO_MID 375 // 1.5ms → 90°
                #define SERVO_MAX 600 // 2.5ms → 180°

                // ── Channel assignments ─────────────────────────
                #define CH_NECK_PAN 0
                #define CH_NECK_TILT 1
                #define CH_LSHOULDER 2
                #define CH_LELBOW 3
                #define CH_LGRIPPER 4
                #define CH_RSHOULDER 5
                #define CH_RELBOW 6
                #define CH_RGRIPPER 7
                #define CH_LHIP_ROLL 8
                #define CH_RHIP_ROLL 9
                #define CH_LHIP_PITCH 10
                #define CH_LKNEE 11
                #define CH_LANKLE 12
                #define CH_RHIP_PITCH 13
                #define CH_RKNEE 14
                #define CH_RANKLE 15

                Adafruit_PWMServoDriver pca = Adafruit_PWMServoDriver(0x40);
                Adafruit_MPU6050 mpu;
                Adafruit_SSD1306 oled(128, 64, &Wire);
                WebSocketsServer ws(81);

                // ── Helper: set servo by angle ──────────────────
                void setServo(uint8_t ch, int angleDeg) {
                angleDeg = constrain(angleDeg, 0, 180);
                int pulse = map(angleDeg, 0, 180, SERVO_MIN, SERVO_MAX);
                pca.setPWM(ch, 0, pulse);
                }

                // ── Init neutral stance (all at 90°) ────────────
                void standNeutral() {
                for (int i = 0; i < 16; i++) setServo(i, 90);
                }

                // ── Cliff detection (returns true if safe) ──────
                bool isSafeGround() {
                return digitalRead(4) && digitalRead(13);
                }

                // ── VC-02 voice command handler ─────────────────
                void handleVoiceCommand(byte
                cmd) {
                switch(cmd) {
                case 0x01: standNeutral();
                break; // "Stand up"
                case 0x02: /* walkForward() */
                break;
                case 0x03: /* danceMode() */
                break;
                case 0x04: /* grabObject() */
                break;
                }
                }

                void setup() {
                Serial.begin(115200);
                Serial2.begin(115200); // VC-02 UART
                Wire.begin(21, 22); // SDA=21,
                    SCL=22

                pca.begin(); pca.setPWMFreq(50);
                mpu.begin();
                oled.begin(SSD1306_SWITCHCAPVCC, 0x3C);

                standNeutral(); // Always calibrate first
                delay(1000);

                WiFi.begin("YourSSID", "YourPassword");
                ws.begin(); ws.onEvent(wsHandler);
                }

                void loop() {
                ws.loop();
                if (Serial2.available()) {
                byte cmd = Serial2.read();
                handleVoiceCommand(cmd);
                }
                if (!isSafeGround()) {
                standNeutral(); // Emergency stop on cliff
                }
                }
            

✅

Test order: (1) Compile blank sketch — check board detected. (2) Add PCA9685 scan — check I2C address 0x40 found. (3) Add one servo sweep on ch0. (4) Add MPU6050 — check serial output. (5) Add VC-02 UART echo. Never add all code at once.

08

Assembly sequence — step by step

🔴

Golden rule: ALWAYS test code before cutting cardboard for each new part. If software works → build hardware. If hardware works → glue permanently. Never glue and then test.

Day 1–2: Breadboard test (no cardboard)

Install Arduino IDE, add ESP32 board package URL
Connect ESP32-CAM via USB (use USB-to-TTL adapter)
Upload blink sketch → LED blinks = success
Wire PCA9685 to ESP32 (SDA=21, SCL=22, 5V, GND)
Run I2C scanner sketch → confirm 0x40 found
Wire ONE MG90S to CH0 → run sweep code
Wire MPU6050 → confirm accelerometer values in serial
Wire OLED → display "Jarvis Boot" text
Connect VC-02 → speak "jarvis" → get serial byte response

Day 3–4: Head build

Cut head box (6×5×4cm) — 3-layer cardboard
Cut eye holes (18mm circles) — left and right
Insert dummy lens in left eye (marble/glass)
Mount OLED at bottom-center front of head box
Mount ESP32-CAM inside — lens aligned to right eye hole
Cut L-bracket for neck (3-layer cardboard)
Code test: run OLED talking wave animation
Mount pitch MG90S in L-bracket horizontal arm
Mount yaw MG90S in torso top (don't build torso yet)
Attach head box to pitch servo horn — test pan+tilt

Day 5–6: Torso + voice

Cut torso box (9×8×5cm) — 4-layer cardboard
Foam-tape all boards inside: PCA9685, MB102, ESP32-CAM
Drill cable holes: I2C, UART, power
Mount VC-02 speaker grille cutout on front
Connect all I2C devices on shared bus (SDA/SCL)
Power test: MB102 → 5V → check all chips boot
Full voice test: speak all 4 commands, verify serial codes
Attach neck + head to torso top

Day 7–8: Arms

Calibrate CH2–CH7 all to 90° via code — confirm in serial
Cut upper arm strips (5cm × 2cm, tin/cardboard)
Cut lower arm strips (5cm × 2cm)
Slot MG90S shoulder (CH2/CH5) on torso sides
Attach upper arm to shoulder horn
Slot MG90S elbow (CH3/CH6) at lower end of upper arm
Attach lower arm to elbow horn
Build gripper: 2 cardboard fingers, pivot point, MG90S CH4/CH7
Test arm wave motion in code

Day 9–11: Hip frame + legs (most complex)

Simulate gait in Webots FIRST before building
Cut hip frame (12×4×5cm) — tin + cardboard hybrid
Slot 2× MG996R Roll servos (CH8, CH9) — shafts face down
Cut 2 thigh pieces (7cm) with Hip Pitch slot + Knee slot
Slot MG996R Pitch (CH10/CH13) at thigh top
Cut shin pieces (7cm) with Ankle slot
Build flat foot plates (10×6cm) — 4-layer thick
Mount IR sensors flat under each foot
Calibrate CH8–CH15 to 90° — robot should stand straight
Add MPU6050 PID balance loop — test static balance
Test cliff detection: move robot near table edge

Day 12+: Walking + AI integration

Implement 4-phase gait cycle in code
Phase 1: shift weight left (hip roll)
Phase 2: lift right leg (hip pitch + knee)
Phase 3: swing right leg forward
Phase 4: place foot, shift weight right
Tune PID constants Kp, Ki, Kd with MPU feedback
Start laptop WebSocket server (Python FastAPI)
Connect Gemini API for voice response generation
Build Next.js dashboard for manual servo control
Add ArUco marker detection for navigation

09

5 build phases — milestones

P1

Head + Voice (Week 1–2)

Robot speaks, listens offline, and shows expressions. No legs or arms yet.

ESP32-CAM detected by Arduino IDE
PCA9685 responds on I2C 0x40
VC-02 decodes "jarvis" and returns command byte via UART
OLED shows talking wave animation on voice command
Head pan + tilt working via 2 MG90S servos
Wi-Fi connected, WebSocket server responding

P2

Arms + Grippers (Week 3)

Robot can wave, extend arms, and grasp objects.

Both arms sweep 0°–180° on command
Gripper opens/closes on voice command "grab"
Dance wave routine working (arm swing animation)
Gemini API integrated for smart responses

P3

Legs — Static Balance (Week 4–5)

Robot stands upright without falling. PID loop working.

All 8 leg servos calibrated at 90° — robot stands straight
MPU6050 tilt data printing to serial at 100Hz+
PID balance loop maintains upright when pushed
Cliff detection stops motion at table edge
Front obstacle detection working

P4

Walking + Navigation (Week 6–8)

Robot walks, turns, and navigates using camera + markers.

4-phase gait cycle walking forward 5+ steps
Turn left/right commands working
ArUco marker detection — robot follows marker
Dance routine: full body coordination (bhangra moves)
Voice commands: walk, turn, dance, stop, grab

P5

AI + Autonomous Tasks (Week 9+)

Robot sees, thinks, and acts. Trash pickup, object tracking, full AI interaction.

Camera sends frame to Python → Gemini Vision → detects objects
Robot autonomously navigates to dustbin and drops wrapper
Face tracking: camera follows your face as you move
Next.js dashboard: live camera feed + servo sliders
Render.com + UptimeRobot: cloud server running 24/7
Full Android app (Flutter) with mobile automation

10

Complete budget breakdown

Item	Qty	Unit price	Total	Source	Phase
AI-Thinker VC-02 Kit (Mic + Speaker)	1	₹700	₹700	ESC Labs	P1
ESP32-CAM with OV2640	1	₹600	₹600	Robu.in	P1
PCA9685 16-channel I2C driver	1	₹220	₹220	Robu.in	P1
MPU6050 6-axis IMU	1	₹110	₹110	Robu.in	P3
0.96" OLED SSD1306 I2C	1	₹200	₹200	Robu.in	P1
MG996R TowerPro Metal Gear (11kg/cm)	4	₹254	₹1,016	Quartz Components	P3
MG90S Metal Gear Micro Servo	12	₹90	₹1,080	Local market / Robu	P1–P3
MB102 dual power supply module	1	₹70	₹70	Robu.in	P1
18650 Li-ion cells (2600mAh) × 2	2	₹110	₹220	Local electronics	P3
TP4056 Type-C charging module	1	₹30	₹30	Local market	P3
IR obstacle sensor (cliff)	2	₹35	₹70	Local market	P3
HC-SR04 ultrasonic	1	₹50	₹50	Local market	P3
Jumper wires (M-M + M-F ribbons)	2 packs	₹50	₹100	Local market	P1
Male + Female header pins (berg strips)	3 strips	₹10	₹30	Local market	P1
General purpose dotted PCB × 3	3	₹15	₹45	Local market	P2
Resistors pack (4.7kΩ, 1kΩ, 2kΩ mix)	1 pack	₹30	₹30	Local market	P1
Black electrical tape	1 roll	₹15	₹15	Local market	P1
Double-sided foam tape	1 roll	₹40	₹40	Local market	P1

Total hardware cost ₹4,626

Tools (one-time investment — not robot cost)

Tool	Price	Where to buy
Soldron 25W soldering iron (original, black handle)	₹250–280	Lajpat Rai Market or Robu.in
XL830L digital multimeter (orange/green rubber cover, buzzer)	₹180–220	Lajpat Rai Market
Bharti/Solderite 60/40 solder wire 50g + Quickfix paste	₹100	Local market
Hot glue gun 20W + 5 glue sticks	₹120	Local hardware shop
Wire cutter + stripper (green/yellow handle)	₹50	Local market
ESD-safe tweezers set (1 straight + 1 curved)	₹45	Local market
Desoldering wick + iron stand	₹70	Local market
Heavy utility cutter (metal body)	₹35	Local hardware shop
Tools total	~₹870

✅

Phase 1 minimum spend: You can start with just ₹1,400 — ESP32-CAM (₹600) + PCA9685 (₹220) + OLED (₹200) + 2 MG90S (₹180) + wires (₹200). Get voice + display + neck working before spending on legs.