Qualcomm Challenge · InterHack BCN Intelligent, connected, and built to keep cities clean.

Smart-Bins is an edge AI system that automates waste classification, deters urban wildlife, and gives sanitation teams real-time visibility into every bin across the city.

Built on the Arduino Uno Q with Qualcomm on-device AI, Smart-Bins tackles two of Barcelona's biggest urban sustainability challenges: incorrect recycling and animal infestations near waste collection points.

A top-mounted camera analyzes waste as it is deposited. A custom-trained AI model classifies it in real time into one of five categories — Glass, Paper/Cardboard, Plastic/Packaging, Organic, or General Waste — and mechanically redirects it to the correct bag.

The same camera monitors the surrounding perimeter at ground level. When a rat, pigeon, or wild boar is detected, the system activates LED strobes and acoustic deterrents to scare off the animal, safely and without harm.

A real-time web platform gives sanitation teams full visibility into every bin: fill levels by waste type, animal activity alerts, and predictions on when each bin will need to be collected.

┌─────────────────────────────────────────────────┐
│                 Arduino Uno Q                   │
│                                                 │
│  ┌─────────────┐       ┌─────────────────────┐  │
│  │  STM32U585  │◄─────►│  Qualcomm QRB2210   │  │
│  │    (MCU)    │ Bridge│  Linux + AI Runtime │  │
│  │  C++ Sketch │       │  Python App         │  │
│  └──────┬──────┘       └──────────┬──────────┘  │
│         │                         │             │
│   Modulino Pixels           Logitech Camera     │
│   Modulino Buzzer           Edge Impulse Model  │
│   Modulino Distance         WebUI Dashboard     │
└─────────────────────────────────────────────────┘
                         │
                    REST API
                         │
              ┌──────────▼──────────┐
              │   Express API        │
              │   Node.js · Railway  │
              └──────────┬──────────┘
                         │
              ┌──────────▼──────────┐
              │   Dashboard          │
              │   Astro + React      │
              │   Vercel             │
              └─────────────────────┘

Lightweight Express server running on Railway. Holds bin state in memory and serves it to the dashboard.

POST /api/update-bin

{
  "id": "1",
  "postDistance": 245
}

GET /api/bins returns the in-memory state of every registered bin including fill level, last update timestamp, and animal incident count.

1. Real-Time Map — Geographic overview of all bins, color-coded: green (empty/mid), yellow (filling), red (critical).
2. Deposit Status — Fill level per waste fraction: Glass, Plastic, Paper, Organic, General.
3. Predictive Analytics — Estimates when each bin will reach capacity based on fill rate. Example: "Will be full at 22:00 — 10h before scheduled collection."
4. Wildlife Incidents — Timestamped log of animal detections with severity levels relative to historical averages.
5. Historical Trends — Per-bin fill rate history used to optimize collection schedules and cut unnecessary pickups.

The detection model runs on Edge Impulse, trained on custom datasets for waste classification and urban pest detection.

Detection uses a 10-frame consensus filter: a label must appear in 10 consecutive frames before triggering any hardware action, eliminating false positives at runtime without retraining.

Camera frame
     │
     ▼
Edge Impulse Model
     │
     ├── Waste label (verde / amarillo / azul)
     │         └── setWasteLed(color) → Modulino Pixels
     │
     └── Animal detected
               └── 10-frame consensus filter
                         └── setAnimalLed(ON) + Buzzer 10s
                                   └── POST /api/update-bin

Distance sensor (every 2s)
     └── Fill level → Dashboard + POST /api/update-bin

• Arduino Uno Q
• arduino-cli + arduino-app-cli
• Python 3.11+
• ADB (Android Debug Bridge)
• Node.js 18+

# Board app
git clone https://github.com/LilDebussy/SmartBins-Frontend
cd smart-bins

# Backend (separate repo)
git clone https://github.com/LilDebussy/SmartBins-Backend
cd smart-bins-backend && npm install && node index.js

adb devices
adb shell
bash

arduino-app-cli app start /home/arduino/ArduinoApps/microwasteanimals

cd sketch
arduino-cli compile --profile default
arduino-cli upload --profile default

adb forward tcp:7000 tcp:7000

Open http://localhost:7000.

smart-bins/
├── sketch/                  # MCU firmware (C++ / Arduino)
│   ├── sketch.ino
│   ├── distance.h           # ToF sensor — fill level
│   ├── leds.h               # 4× Modulino Pixels control
│   ├── buzzer.h             # Acoustic deterrent
│   └── sketch.yaml          # Board profile and dependencies
├── python/                  # MPU application (Python)
│   └── main.py              # AI detections + Bridge + API
├── assets/                  # Web UI (served by WebUI brick)
│   ├── index.html
│   ├── app.js
│   └── style.css
├── app.yaml                 # App Lab configuration
└── README.md

Built at InterHack BCN · Qualcomm Challenge — Race Condition

• Lucia Acedo
• Juan Carlos Díaz
• Xavier Román
• Joan Aranda
• Nil Babot