SMART VEHICLE / OPEN-SOURCE ROBOTICS

Code that sees.
Hardware that moves.

A Raspberry Pi vehicle with computer vision, an Android control station, and a lot going on under the chassis.

SmartVehicle brings YOLO detection, live video, and motor control together in one open-source engineering project. Built to explore what happens when software leaves the screen.

See how it works Explore the code ↗

Raspberry Pi 5 · Python · Kotlin · Docker

Smart Vehicle prototype with a blue enclosure and Mecanum wheels — From the bench to the floor. Made to leave
the desk. ↗

CAMERA → COMPUTE → CONTROL ON-DEVICE VISION. OPEN CODE. REAL-WORLD CONSTRAINTS.

01 / THE PROJECT

A small vehicle.
A whole system to figure out.

Not just an AI model on a board. The interesting work is getting vision, movement, networking, and a usable client to work together.

01 — SEE

Understand the camera feed.

A Pi Camera supplies video to the Raspberry Pi. YOLO detects objects on the vehicle; the Android client handles text recognition with Google ML Kit.

02 — MOVE

Put the controls in your hand.

The native Android app combines a live camera view, telemetry, and Mecanum drive controls. Four independently driven wheels let the platform move in multiple directions.

03 — KEEP GOING

Plan for the messy parts.

Separate services keep heavy vision work away from motor control. A network watchdog provides a fallback hotspot when Wi-Fi disappears.

02 / ENGINEERING NOTES

One vehicle.
Independent moving parts.

The first prototype coupled vision and motor control. When inference got busy, control waited. That constraint shaped the architecture.

INPUT

Pi Camera v2

Video capture · MIPI CSI-2

ON THE VEHICLE

Raspberry Pi 5

YOLO vision · Python services · Docker

Video stream Motor control

IN YOUR HAND

Android control station

Live view · Drive commands · Telemetry · OCR

A / DECOUPLE THE WORK

Vision shouldn’t block movement.

Multithreaded pipelines and Docker services separate camera processing, inference, and control. The goal is to keep a busy model from holding up the rest of the system.

B / KEEP COMPUTE CLOSE

Detection happens on the Pi.

YOLO runs on the vehicle. Video travels to the Android client, where the operator can control the platform and use ML Kit to extract text.

C / EXPECT A DROPPED CONNECTION

Give the network a fallback.

Smart Network monitors connectivity and brings up a local hotspot when the external network drops. Recovery is part of the design.

03 / UNDER THE CHASSIS

Off-the-shelf parts.
Hands-on engineering.

Hardware block diagram connecting power, Raspberry Pi, camera, and motor drivers — FIG. 02 / Power, sensing, and movement — connected.

Compute: Raspberry Pi 5 · 8GB RAM
Camera: Pi Camera Module v2 · 8MP
Drive: 4 DC geared motors · Mecanum wheels
Motor control: L298N H-bridge drivers
Power: XL4015 step-down converter · 5.1V / 5A

04 / FROM THE BUILD

Past the architecture diagram.

Actual interfaces and system captures from the project. Open any image for a closer look.

Android client showing YOLO object detection — 01 / OBJECT DETECTIONWhat the vehicle sees.

Android client showing recognized text with ML Kit OCR — 02 / TEXT RECOGNITIONReading the world around it.

Docker service logs on the Raspberry Pi — 03 / SERVICESUnder the hood, in the logs.

System monitor showing Raspberry Pi CPU and memory usage — 04 / RESOURCESKeeping an eye on the Pi.

05 / OPEN THE SOURCE

No black box.
Follow the code.

Start with the server and Android client, then explore the supporting pieces. Setup instructions live alongside the implementation in each repository.

Python / Docker

The vehicle’s backend

GPIO control, YOLO inference, and MJPEG video streaming on the Raspberry Pi.

rpi-server

Kotlin / Jetpack Compose

The control station

Drive the vehicle, watch its camera feed, inspect telemetry, and read text with ML Kit OCR.

android-client

Bash / Linux

A way back online

A network watchdog that creates a fallback Wi-Fi hotspot when the external network drops.

smart-network

ONNX / TFLite

Models for the edge

Export and quantization workflows for deploying YOLO models on constrained hardware.

model-pipeline

More to explore: Web dashboard ↗ · Original server prototype ↗

This website’s source ↗

06 / WHERE IT STARTED

Built to learn.
Kept building.

Smart Vehicle — originally Pametno Vozilo — grew out of hands-on robotics work at ETŠ “Nikola Tesla” in Niš. It earned 4th place and a Special Award at the 10th Galaksija Cup.

The competition is part of its history. The open-source code and the engineering lessons are what carry it forward.

Created by Danilo Stoletović.
Mentor: Predrag Šubarević.
With thanks to former mentor Dejan Batanjac.

Fourth-place certificate and Special Award from the 10th Galaksija Cup — FROM THE PROJECT ARCHIVE / 10TH GALAKSIJA CUP

07 / WORK IN PROGRESS

A platform to keep building on.

THE FIRST BUILD / V1
Make it move.

Chassis assembly, manual Wi-Fi control, a camera feed, and early YOLO experiments.
THE FOUNDATION / V2
Make the pieces work together.

Docker services, a Compose Android client, YOLO and OCR, and fallback networking.
FUTURE DIRECTIONS / V3
Explore more autonomy.

LiDAR and SLAM, accessible programming tools, and clearer replication guides. These are planned directions, not shipped capabilities.

TAKE IT APART. UNDERSTAND IT. BUILD ON IT.

Your next experiment
could have wheels.

Explore the implementation, reproduce a part of the system, or bring an improvement back to the project. The repositories are the place to start.

Explore SmartVehicle on GitHub ↗ Meet the person behind it ↗

A practical starting point for studying computer vision, embedded control, and connected systems.

Code that sees.Hardware that moves.

A small vehicle.A whole system to figure out.