Task-agnostic evolution of diverse repertoires of swarm behaviours

Jorge Gomes*, Anders Lyhne Christensen

*Kontaktforfatter for dette arbejde

Publikation: Kapitel i bog/rapport/konference-proceedingKonferencebidrag i proceedingsForskningpeer review

Abstrakt

Quality diversity algorithms are evolutionary algorithms that aim to evolve diverse repertoires of high-quality solutions. Quality diversity has recently been used with considerable success to evolve repertoires of single-robot controllers in a wide range of applications. In this paper, we propose a methodology for the evolution of repertoires of general swarm behaviours. We use a quality diversity algorithm that relies on a behaviour characterisation and a quality metric that are task-agnostic, meaning that the repertoire evolution is not driven towards solving any specific task. We use a total of eight swarm robotics tasks to evaluate the behaviours contained in the evolved repertoires a-posteriori, and compare their performance with direct task-specific evolution. We show that the repertoires contain a wide diversity of swarm behaviours, and for most of the tasks, the behaviours in the repertoire have a performance close to the performance achieved by task-specific evolution.

OriginalsprogEngelsk
TitelSwarm Intelligence : 11th International Conference, ANTS 2018, Proceedings
RedaktørerMarco Dorigo, Mauro Birattari, Christian Blum, Anders L. Christensen, Andreagiovanni Reina, Vito Trianni
ForlagSpringer VS
Publikationsdatookt. 2018
Sider225-238
ISBN (Trykt)978-3-030-00532-0
ISBN (Elektronisk)978-3-030-00533-7
DOI
StatusUdgivet - okt. 2018
Begivenhed11th International Conference on Swarm Intelligence, ANTS 2018 - Rome, Italien
Varighed: 29. okt. 201831. okt. 2018

Konference

Konference11th International Conference on Swarm Intelligence, ANTS 2018
LandItalien
ByRome
Periode29/10/201831/10/2018
NavnLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Vol/bind11172 LNCS
ISSN0302-9743
NavnTheoretical Computer Science and General Issues
Vol/bind11172

Fingeraftryk

Dyk ned i forskningsemnerne om 'Task-agnostic evolution of diverse repertoires of swarm behaviours'. Sammen danner de et unikt fingeraftryk.

Citationsformater