A learnheuristic approach for the team orienteering problem with aerial drone motion constraints

This work proposes a learnheuristic approach (combination of heuristics with machine learning) to solve an aerial-drone team orienteering problem. The goal is to maximise the total reward collected from information gathering or surveillance observations of a set of known targets within a fixed amoun...

Descripción completa

Detalles Bibliográficos
Autores: Bayliss, Christopher|||0000-0003-0031-5937, Juan, Ángel A.|||0000-0003-1392-1776, Currie, Christine S.M., Panadero, Javier|||0000-0002-3793-3328
Tipo de recurso: artículo
Fecha de publicación:2020
País:España
Institución:Universitat Autònoma de Barcelona
Repositorio:Dipòsit Digital de Documents de la UAB
Idioma:inglés
OAI Identifier:oai:ddd.uab.cat:296841
Acceso en línea:https://ddd.uab.cat/record/296841
https://dx.doi.org/urn:doi:10.1016/j.asoc.2020.106280
Access Level:acceso abierto
Palabra clave:Aerial drones
Learnheuristics
Machine learning
Metaheuristics
Route-dependent edge times
Team orienteering problem
Descripción
Sumario:This work proposes a learnheuristic approach (combination of heuristics with machine learning) to solve an aerial-drone team orienteering problem. The goal is to maximise the total reward collected from information gathering or surveillance observations of a set of known targets within a fixed amount of time. The aerial drone team orienteering problem has the complicating feature that the travel times between targets depend on a drone's flight path between previous targets. This path-dependence is caused by the aerial surveillance drones flying under the influence of air-resistance, gravity, and the laws of motion. Sharp turns slow drones down and the angle of ascent and air-resistance influence the acceleration a drone is capable of. The route dependence of inter-target travel times motivates the consideration of a learnheuristic approach, in which the prediction of travel times is outsourced to a machine learning algorithm. This work proposes an instance-based learning algorithm with interpolated predictions as the learning module. We show that a learnheuristic approach can lead to higher quality solutions in a shorter amount of time than those generated from an equivalent metaheuristic algorithm, an effect attributed to the search-diversity enhancing consequence of the online learning process.