Path and obstacle detection vehicle using YOLO

In this repository, you will find the resources used in the article “A neural network approach to path-detection and self-driving vehicles using YOLO and one-layer neuro-adaptive control”, published in 2025 in the journal Engineering Applications of Artificial Intelligence.

The work presents the development and implementation—both in simulation and in real-time—of a navigation and obstacle detection system based on a 4-wheeled omnidirectional mobile robot. The system integrates computer vision techniques with deep and shallow artificial neural networks, enabling the robot to autonomously detect and follow paths of arbitrary shape (closed circuits or straight lines) and to identify obstacles using an onboard camera.

Path detection is achieved through the training of two YOLO (You Only Look Once) models, while obstacle detection relies on the Tiny-YOLOv2 state-of-the-art architecture. Additional techniques such as transfer learning, coordinate projection, and trajectory generation algorithms are employed to enhance system performance.

Furthermore, a single-layer neuro-adaptive compensation control based on filtered error is implemented to regulate the wheel velocities of the omnidirectional robot. This neural controller compensates for unknown nonlinear dynamics of the robot, with network weights estimated online using appropriate adaptive update laws.

This repository is intended to support reproducibility and further research in autonomous mobile robotics, vision-based navigation, and neuro-adaptive control systems.