FollowThePathNet: UAVs Use Neural Networks to Follow Paths in GPS-Denied Environments

FollowThePathNet: UAVs Use Neural Networks to Follow Paths in GPS-Denied Environments

Paper in proceedings (conference paper)

Navigating complex pathways autonomously poses a significant challenge for Unmanned Aerial Vehicles (UAVs). To address this issue, we developed a robust convolutional neural network (CNN) enabling UAVs to follow specific paths, such as trail, rural, and cycling ones, using real-time camera data. Our CNN model interprets the visual data to estimate the UAV's position relatively to the path, enabling path following without human intervention. This article details the methodology employed in training our neural network, including the data collection, architecture of the model, and parameters. Additionally, we describe integrating the hardware and software components used in the implementation. We conducted real-world tests to evaluate the effectivity of our approach. These tests confirmed the UAVs' capability to follow the designated paths, demonstrating the practical applicability and reliability of the system. The results and their implications are discussed thoroughly.

Navigating complex pathways autonomously poses a significant challenge for Unmanned Aerial Vehicles (UAVs). To address this issue, we developed a robust convolutional neural network (CNN) enabling UAVs to follow specific paths, such as trail, rural, and cycling ones, using real-time camera data. Our CNN model interprets the visual data to estimate the UAV's position relatively to the path, enabling path following without human intervention. This article details the methodology employed in training our neural network, including the data collection, architecture of the model, and parameters. Additionally, we describe integrating the hardware and software components used in the implementation. We conducted real-world tests to evaluate the effectivity of our approach. These tests confirmed the UAVs' capability to follow the designated paths, demonstrating the practical applicability and reliability of the system. The results and their implications are discussed thoroughly.

Training, Visualization, Navigation, Neural networks, Reliability engineering, Data models, Software

Training, Visualization, Navigation, Neural networks, Reliability engineering, Data models, Software

RAICHL, P.; JANOUŠEK, J.; MARCOŇ, P.

2025

19.06.2024

IEEE

New York

979-8-3503-5788-2

2024 International Conference on Unmanned Aircraft Systems

2575-7296

2024 International Conference on Unmanned Aircraft Systems (ICUAS)

Hellenic Republic

92

98

7

https://ieeexplore.ieee.org/document/10557044/keywords#keywords

http://hdl.handle.net/