Combining the A* Algorithm with Neural Networks to Solve the Team Orienteering Problem with Obstacles and Environmental Factors

This paper addresses the team orienteering problem applied to unmanned aerial vehicles (UAVs), considering obstacle avoidance and environmental factors such as wind conditions and payload weight. The objective is to optimize UAV routes to maximize collected rewards while adhering to operational cons...

ver descrição completa

Detalhes bibliográficos
Autores: Freixes, Alfons|||0000-0002-1675-358X, Panadero, Javier|||0000-0002-3793-3328, Juan, Ángel A.|||0000-0003-1392-1776, Serrat, Carles|||0000-0002-1504-5354
Formato: artículo
Fecha de publicación:2025
País:España
Recursos:Universitat Autònoma de Barcelona
Repositorio:Dipòsit Digital de Documents de la UAB
Idioma:inglés
OAI Identifier:oai:ddd.uab.cat:319619
Acesso em linha:https://ddd.uab.cat/record/319619
https://dx.doi.org/urn:doi:10.3390/a18060309
Access Level:acceso abierto
Palavra-chave:A* algorithm
Artificial intelligence
Team orienteering problem
Unmanned aerial vehicles
Descrição
Resumo:This paper addresses the team orienteering problem applied to unmanned aerial vehicles (UAVs), considering obstacle avoidance and environmental factors such as wind conditions and payload weight. The objective is to optimize UAV routes to maximize collected rewards while adhering to operational constraints. To achieve this, we employ a simheuristic algorithm for the overall route optimization, while integrating the A* algorithm to determine feasible paths between nodes that avoid obstacles in a 2D grid-based environment. Then, a feedforward neural network estimates travel time based on UAV speed, wind conditions, trajectory distance, and payload weight. This estimation is incorporated into the optimization process to improve route planning accuracy. Numerical experiments evaluate the impact of various parameters, including obstacle placement, UAV speed, wind conditions, and payload weight. These experiments include maps with 30 to 100 points of interest and varying obstacle densities and show that our hybrid method improves solution quality by up to (Formula presented.) in total profit compared to a baseline approach. Furthermore, computation times remain within 5-10% of the baseline, showing that the added predictive layer maintains computational efficiency.