🌟 Overview

GCBOT is a remotely controlled garbage-collecting robot built for competitions and real-world waste management challenges. It combines:

• 🏎️ A 4-wheel drive chassis controlled via a motor driver
• 🦾 A servo-powered arm & gripper for picking up trash
• 📷 A live camera stream from an onboard Raspberry Pi
• 🧠 YOLOv10 AI running real-time trash detection on the laptop
• 🏆 Combined AI + Weight scoring — points are based on both what the AI detected AND how heavy the object is
• ⚖️ HX711 load cell measures object weight for score multiplier
• 🌐 A browser-based control panel accessible from any device on the network
• 🔐 Login & player system — sign in or create an account before controlling the robot
• 📲 QR onboarding page — scan to join Wi-Fi, scan again to open the game instantly

The entire system communicates over TCP sockets — the laptop runs the Flask web server, AI inference, and score computation; the Raspberry Pi acts as a bridge between the laptop and the Arduino; and the Arduino directly controls all motors, servos, and the load cell.

📸 Screenshots

🔐 Login Page

Players sign in or create an account before accessing the robot controls. The glassmorphism login card sits on the animated dark background.

📲 QR Onboarding Page (/qr)

Displayed on a screen at the competition venue. Players scan Step 1 to join the Wi-Fi hotspot, then scan Step 2 to instantly open the control page — no typing required.

📱 Mobile Views

The interface is fully responsive. On portrait mobile it prompts rotation; on landscape it shows the full control panel.

🆕 Create Account Page

New players fill in their Full Name, Nickname (shown on leaderboard), choose an avatar, and set a password. The profile is saved and they are logged in immediately.

🏆 Leaderboard Page

Shows all players ranked by total score. Each entry displays the player's avatar, nickname, full name, and points earned in the session.

🖥️ Web Control Interface (Desktop)

The full-screen landscape control panel with live AI Detection feed, movement D-pad, gripper & arm controls, and the real-time score badge.

🏗️ System Architecture

┌──────────────────────────────────────────────────────────────────┐
│                   GCBOT — Gamified Cleaning Bot                  │
│                                                                  │
│  ┌────────────┐   WiFi/TCP   ┌──────────────┐   Serial/USB       │
│  │   LAPTOP   │◄────────────►│ Raspberry Pi │◄─────────────────► │
│  │            │              │              │                    │
│  │ Flask App  │              │ pi_control   │   ┌─────────────┐  │
│  │ (Port 5001)│              │ (Port 5000)  │   │   Arduino   │  │
│  │            │              │              │   │             │  │
│  │ YOLOv10    │◄── Video ───►│ Camera       │   │ Motor Driver│  │
│  │ Inference  │   (Port 9999)│ Stream Tx    │   │ Servos (x4) │  │
│  │            │              │              │   │ Load Cell   │  │
│  │ Web UI     │◄── Cmds ────►│ Serial Relay │   │ HX711       │  │
│  └────────────┘              └──────────────┘   └─────────────┘  │
│        ▲                                                         │
│        │  HTTP (Browser)                                         │
│  ┌─────┴──────┐                                                  │
│  │ Any Device │ (Phone / Tablet / PC on same network)            │
│  └────────────┘                                                  │
└──────────────────────────────────────────────────────────────────┘

Three-Tier Architecture Diagram

graph TD
    %% Define Node Styles
    classDef component fill:#ffffff,stroke:#333333,stroke-width:1px;

    subgraph Tier3 ["Tier 3: Host / Fog Layer"]
        Browser[("Web Browser (Mobile/PC)\nControl UI & Video")]:::component
        Flask["Flask Web Server"]:::component
        YOLO["YOLOv10 Inference Engine"]:::component

        Browser <==>|HTTP / WebSocket| Flask
        Flask <==>|Frames / Classes| YOLO
    end
    style Tier3 fill:#e3f2fd,stroke:#1565c0,stroke-width:2px;

    subgraph Tier2 ["Tier 2: Edge Layer"]
        Pi["Raspberry Pi 4"]:::component
        Cam[("USB Web Camera")]:::component

        Cam -->|Raw Video| Pi
    end
    style Tier2 fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px;

    subgraph Tier1 ["Tier 1: Hardware Layer"]
        Arduino["Arduino Uno"]:::component
        L298N["L298N Motor Driver"]:::component
        Motors(("4x DC Motors")):::component
        Servos(("4x Servo Motors\n(Arm & Gripper)")):::component
        HX711["HX711 Amplifier"]:::component
        LoadCell[("1Kg Load Cell")]:::component

        Arduino -->|PWM Signals| L298N
        L298N -->|Power| Motors
        Arduino -->|PWM Signals| Servos
        HX711 -->|Digital Weight Data| Arduino
        LoadCell -->|Analog Signal| HX711
    end
    style Tier1 fill:#fff3e0,stroke:#e65100,stroke-width:2px;

    %% Cross-layer connections
    Flask ==>|TCP Control Commands| Pi
    Pi ==>|MJPEG Video Stream| Flask
    Pi <==>|Serial USB| Arduino

Communication Flow

sequenceDiagram
    participant Browser as 🌐 Browser
    participant Laptop as 💻 Laptop (Flask)
    participant Pi as 🍓 Raspberry Pi
    participant Arduino as ⚡ Arduino

    Browser->>Laptop: HTTP POST /cmd {"cmd": "F"}
    Laptop->>Pi: TCP "F\n" (Port 5000)
    Pi->>Arduino: Serial "F\n"
    Arduino-->>Arduino: forward()

    Browser->>Laptop: HTTP GET /video_feed
    Pi-->>Laptop: TCP Video Frame (Port 9999)
    Laptop-->>Laptop: YOLOv10 Inference → tracks last_detected_class
    Laptop-->>Browser: MJPEG Stream

    Note over Browser,Arduino: DROP OBJECT Flow (AI + Weight Scoring)
    Browser->>Laptop: POST /cmd {"cmd": "WEIGHTNOW"}
    Laptop->>Laptop: weighnow_pending = True
    Laptop->>Pi: TCP "WEIGHTNOW\n"
    Pi->>Arduino: Serial "WEIGHTNOW\n"
    Arduino-->>Arduino: HX711 read → diff = current - lastWeight
    Arduino-->>Pi: Serial "W:23.45\n"
    Pi-->>Laptop: TCP "W:23.45\n" (broadcast)
    Laptop-->>Laptop: compute_combined_score(23.45g)
    Note over Laptop: score = round(base × (1 + 23.45/100))
    Laptop-->>Browser: JSON {score: 12, event: {cls: "Bottle", pts: 12, weight: 23.5}}

✨ Features

🔧 Hardware

Components List

Wiring — Arduino Pin Map

Arduino Uno
├── Pin  2  → L298N IN1  (Left Motor A)
├── Pin  3  → L298N IN2  (Left Motor B)
├── Pin  4  → L298N IN3  (Right Motor A)
├── Pin  7  → L298N IN4  (Right Motor B)
├── Pin  8  → L298N ENA  (Left Motor PWM)
├── Pin 12  → L298N ENB  (Right Motor PWM)
├── Pin  5  → Gripper Servo A (grabA)
├── Pin  6  → Gripper Servo B (grabB) [mirrored]
├── Pin  9  → Lift Servo A (liftA)
├── Pin 10  → Lift Servo B (liftB) [mirrored]
├── Pin A0  → HX711 SCK
└── Pin A1  → HX711 DT

💻 Software Stack

🧠 AI Model — YOLOv10 Trash Detection

The onboard AI model is a custom-trained YOLOv10 fine-tuned on a multi-class garbage detection dataset. It runs locally on the laptop CPU in a dedicated inference thread at 416×416 resolution.

Detected Waste Classes (18 classes)

Aluminium foil · Bottle · Bottle cap · Broken glass · Can · Carton · Cigarette · Cup · Lid · Other litter · Other plastic · Paper · Plastic bag/wrapper · Plastic container · Pop tab · Straw · Styrofoam piece · Unlabeled litter

Training Performance Graphs