Here is the complete step-by-step engineering documentation and implementation guide for building an AI-Based Trash Segregation Robotic Arm using ESP32, Computer Vision, n8n Automation, Telegram Voice Alerts, and Cloud Analytics.
1. Full Project Description
This system automates waste segregation using Edge/Cloud AI and Agentic IoT workflow automation.
-
Computer Vision & Classification: An ESP32-CAM or USB Camera captures images of waste arriving on a conveyor or platform. The frame is evaluated using a lightweight Object Detection Model (YOLOv8 / Teachable Machine model deployed locally or via API).
-
Actuation Mechanics: Based on the detected class (Biodegradable, Non-Biodegradable / Plastic, Metal, Hazardous), the ESP32 micro-controller drives a multi-DOF servo robotic arm to pick and place the item into its respective bin.
-
Agentic IoT & n8n Workflow Automation: Upon each classification, event telemetry is dispatched via HTTP POST/MQTT to an n8n self-hosted instance. An Agentic AI node evaluates continuous operational logs.
-
Cloud Logging & Visualization:
-
ThingSpeak: Logs real-time sensor parameters (current, servo angles, object counts, operational power).
-
Google Sheets: Acts as a relational event database tracking historical sorting records and timestamps.
-
-
Telegram Voice Alerts: n8n converts critical alerts (e.g., bin overflow, motor stall, high power consumption) to speech using TTS (ElevenLabs / OpenAI TTS) and sends
.oggvoice notes to a Telegram group.
2. Components List
Hardware Components
-
ESP32 DevKit V1 (Main Microcontroller)
-
ESP32-CAM Module (Vision capture node)
-
4-DOF or 6-DOF Robotic Arm Kit (with MG996R or SG90 Servos)
-
PCA9685 16-Channel 12-bit PWM Servo Driver (I2C interface)
-
ACS712 Current Sensor (5A) (For power & load monitoring)
-
HC-SR04 Ultrasonic Sensors x3 (Bin full level detection)
-
5V 5A High-Current DC Power Supply (Dedicated for servos)
-
5V 2A Micro-USB Supply (For ESP32 boards)
-
Logic Level Shifter (3.3V to 5V) (Optional, for sensor safety)
Software & Cloud Stack
-
Arduino IDE / PlatformIO (ESP32 Firmware)
-
n8n (Self-hosted workflow automation platform)
-
ThingSpeak API (Telemetry dashboard)
-
Google Sheets API (Data logging)
-
Telegram Bot API (Voice and text alerts)
-
OpenCV / Roboflow / Teachable Machine / Ollama (Image Classification Agent)
3. Circuit Schematic Diagram
Hardware Wiring Connections
| Component | Pin / Terminal | ESP32 Board |
| PCA9685 Servo Driver | VCC | 3.3V / 5V |
| GND | GND | |
| SDA | GPIO 21 | |
| SCL | GPIO 22 | |
| ACS712 Current Sensor | VCC | 5V |
| GND | GND | |
| OUT | GPIO 34 (Analog In) | |
| Ultrasonic Sensor (Bin 1) | Trig / Echo | GPIO 12 / GPIO 13 |
| Ultrasonic Sensor (Bin 2) | Trig / Echo | GPIO 14 / GPIO 27 |
| Servos (Base, Shoulder, Elbow, Gripper) | Channel 0 - 3 | Wired to PCA9685 |
| External Power (5V 5A) | V+ / V- | PCA9685 Power Screw Terminal |
Crucial Power Rule: NEVER power the servo motors directly from the ESP32 5V/3.3V pins. Connect external 5V 5A directly to the PCA9685 terminal block, and ensure a common ground between the external power supply and the ESP32.
4. Flowchart
[Start] --> [ESP32-CAM Captures Image]
|
v
[Send Image to AI Agent / Classifier]
|
v
[Object Class Identified?]
/ | \
(Plastic) (Metal) (Bio-degradable)
/ | \
[Bin A Pos] [Bin B Pos] [Bin C Pos]
\ | /
v
[PCA9685 Drives Servos to Place Item]
|
v
[ACS712 Reads Current + Power Consumption]
|
v
[HTTP POST Payload sent to n8n Webhook]
|
+-------+-------+
| |
v v
[Google Sheets] [ThingSpeak]
(Row Added) (Fields Updated)
|
v
[n8n Agentic AI Evaluates Metrics]
|
{Is Bin Full OR High Power Spike?}
/ \
(Yes) (No)
/ \
[Generate TTS Voice] [End Cycle]
|
[Telegram Voice Alert]
5. ESP32 Source Code
Flash this C++ code onto your main ESP32 DevKit V1 using the Arduino IDE. Make sure to install the Adafruit_PWMServoDriver and WiFi libraries.
#include <WiFi.h>
#include <HTTPClient.h>
#include <Wire.h>
#include <Adafruit_PWMServoDriver.h>
// WiFi Configuration
const char* ssid = "YOUR_WIFI_SSID";
const char* password = "YOUR_WIFI_PASSWORD";
// n8n Webhook Endpoint
const char* n8n_webhook_url = "http://YOUR_N8N_IP:5678/webhook/trash-segregation";
// PCA9685 PWM Setup
Adafruit_PWMServoDriver pwm = Adafruit_PWMServoDriver();
#define SERVOMIN 150 // Minimum pulse length count out of 4096
#define SERVOMAX 600 // Maximum pulse length count out of 4096
// Sensor Pins
#define CURRENT_SENSOR_PIN 34
void setup() {
Serial.begin(115200);
Wire.begin(21, 22); // SDA, SCL
pwm.begin();
pwm.setPWMFreq(60); // Analog servos run at ~60Hz
WiFi.begin(ssid, password);
while (WiFi.status() != WL_CONNECTED) {
delay(500);
Serial.print(".");
}
Serial.println("\nWiFi Connected!");
// Set default home position
moveHome();
}
void loop() {
if (Serial.available() > 0) {
String input = Serial.readStringUntil('\n');
input.trim();
if (input.startsWith("SORT:")) {
String trashType = input.substring(5);
float currentmA = readCurrent();
float powerWatts = (currentmA / 1000.0) * 5.0; // P = V * I
executeSorting(trashType);
sendTelemetryToN8N(trashType, currentmA, powerWatts);
}
}
}
void setAngle(uint8_t num, double angle) {
double pulse = map(angle, 0, 180, SERVOMIN, SERVOMAX);
pwm.setPWM(num, 0, pulse);
}
void moveHome() {
setAngle(0, 90); // Base
setAngle(1, 45); // Shoulder
setAngle(2, 45); // Elbow
setAngle(3, 0); // Gripper Open
}
void executeSorting(String category) {
// Pick sequence
setAngle(3, 90); // Close Gripper
delay(500);
setAngle(1, 90); // Lift arm
delay(500);
// Rotate base to designated bin
if (category == "PLASTIC") setAngle(0, 30);
else if (category == "METAL") setAngle(0, 90);
else if (category == "ORGANIC") setAngle(0, 150);
else setAngle(0, 180); // Default/Unknown
delay(1000);
setAngle(3, 0); // Drop item
delay(500);
moveHome(); // Return to ready state
}
float readCurrent() {
int rawADC = analogRead(CURRENT_SENSOR_PIN);
float voltage = (rawADC / 4095.0) * 3.3;
// Offset for ACS712 5A module (VCC/2 centered, ~185mV/A sensitivity)
float currentA = (voltage - 1.65) / 0.185;
return abs(currentA * 1000.0); // Return mA
}
void sendTelemetryToN8N(String type, float current, float power) {
if (WiFi.status() == WL_CONNECTED) {
HTTPClient http;
http.begin(n8n_webhook_url);
http.addHeader("Content-Type", "application/json");
String jsonPayload = "{";
jsonPayload += "\"waste_type\":\"" + type + "\",";
jsonPayload += "\"current_mA\":" + String(current) + ",";
jsonPayload += "\"power_W\":" + String(power);
jsonPayload += "}";
int httpResponseCode = http.POST(jsonPayload);
Serial.print("n8n Webhook Response: ");
Serial.println(httpResponseCode);
http.end();
}
}
6. n8n Workflow JSON
Save the block below as a .json file and import it directly into your n8n canvas via Workflow -> Import from File.
{
"name": "AI Trash Segregation & Alert Workflow",
"nodes": [
{
"parameters": {
"httpMethod": "POST",
"path": "trash-segregation",
"options": {}
},
"id": "node-1",
"name": "ESP32 Webhook",
"type": "n8n-nodes-base.webhook",
"typeVersion": 1,
"position": [250, 300]
},
{
"parameters": {
"operation": "append",
"sheetId": "YOUR_GOOGLE_SHEET_ID",
"range": "Sheet1!A:C",
"options": {}
},
"id": "node-2",
"name": "Google Sheets",
"type": "n8n-nodes-base.googleSheets",
"typeVersion": 4,
"position": [500, 200]
},
{
"parameters": {
"requestMethod": "GET",
"url": "=https://api.thingspeak.com/update?api_key=YOUR_THINGSPEAK_WRITE_KEY&field1={{$json.body.current_mA}}&field2={{$json.body.power_W}}"
},
"id": "node-3",
"name": "ThingSpeak Update",
"type": "n8n-nodes-base.httpRequest",
"typeVersion": 4.1,
"position": [500, 400]
},
{
"parameters": {
"conditions": {
"number": [
{
"value1": "={{ $json.body.power_W }}",
"operation": "larger",
"value2": 2.5
}
]
}
},
"id": "node-4",
"name": "Check Anomaly",
"type": "n8n-nodes-base.if",
"typeVersion": 1,
"position": [720, 300]
},
{
"parameters": {
"chatId": "YOUR_TELEGRAM_CHAT_ID",
"text": "=⚠️ ALERT: High Servo Power Consumption Detected! Voltage Draw: {{ $json.body.power_W }} W. Check for motor jam."
},
"id": "node-5",
"name": "Telegram Alert",
"type": "n8n-nodes-base.telegram",
"typeVersion": 1,
"position": [950, 200]
}
],
"connections": {
"ESP32 Webhook": {
"main": [
[
{ "node": "Google Sheets", "type": "main", "index": 0 },
{ "node": "ThingSpeak Update", "type": "main", "index": 0 },
{ "node": "Check Anomaly", "type": "main", "index": 0 }
]
]
},
"Check Anomaly": {
"main": [
[
{ "node": "Telegram Alert", "type": "main", "index": 0 }
]
]
}
}
}
7. Setup & Configurations
Search for
@BotFatherin Telegram and start a chat.Send
/newbot, name your bot, and save the generated HTTP API Token.Search for
@userinfobot, press/start, and copy your personal Id (Chat ID).Paste these credentials into the n8n Telegram Node settings.
Go to Google Cloud Console and enable the Google Sheets API.
Create a Service Account, download the JSON key file, and link it inside n8n under OAuth2/Service Account credentials.
Create a Google Sheet with headers:
Timestamp | Waste Category | Current (mA) | Power (W).Share the sheet with the Service Account email address giving Editor permission.
Sign up at ThingSpeak.com.
Create a New Channel named
AI Trash Segregation System.Enable two fields: Field 1:
Current (mA), Field 2:Power (Watts).Copy the Write API Key and paste it into the n8n HTTP Request node URL.
The ACS712 sensor streams current data. If a servo gets jammed by heavy waste:
Standard operational draw = 0.2A to 0.5A (1.0W - 2.5W).
Stall current = > 1.2A (> 6.0W).
The n8n agent detects values over threshold and dynamically routes an urgent speech warning to Telegram.
Inside n8n, send the anomaly prompt text to an OpenAI TTS node (
tts-1model) or ElevenLabs API.Set output audio encoding to
.ogg/OPUS.Wire the resulting audio binary to the Telegram Node (Send Audio / Voice Note).
8. Future Enhancements & Deployment Guide
-
Edge AI Processing: Replace cloud-based image inference with Edge Impulse / ESP32-S3 Eye or a Raspberry Pi 4/5 running local YOLOv8-nano to remove latency completely.
-
Reverse Kinematics: Upgrade from hardcoded angular movements to inverse kinematics (IK) trajectories for smooth motion profiles and adaptive grip height.
-
Solar & Battery System: Add an 18650 Li-ion battery backup array with solar charging for off-grid municipal deployment.
No comments:
Post a Comment