Body and brain quality-diversity in robot swarms

Mkhatshwa, Sindiso

Body and brain quality-diversity in robot swarms

Thesis / Dissertation

2023

Publisher

University of Cape Town

Department

Department of Computer Science

Faculty

Faculty of Science

Abstract

Various studies have shown that diverse groups perform better, solve problems more adeptly, and are more resilient. However, in evolutionary robotics, evolving group diversity is a difficult task that frequently calls for geographic isolation, a division of labor mechanism, and a careful choice of parameters. According to recent research, decentralized Quality Diversity (QD) algorithms can generate behavioral diversity across a swarm without requiring geographical isolation or a division of labor mechanism. Despite the fact that these findings represent an essential first step in the quest to find a mechanism to evolve behavioral diversity across a swarm in physical robot tasks, little research has been done on evolving behavior-morphology diversity across a robot swarm given cooperative tasks. To address this issue, we investigate the application of a decentralized QD algorithm (EDQD) to generate group diversity given an increasingly challenging collective behavior task in order to determine the circumstances in which it succeeds and fails. We further develop Double-Map EDQD-M, an algorithm that combines morphology characterization and behavior characterization (body-brain diversity maintenance). Results indicate that body-brain diversity maintenance yielded significantly higher behavioral and morphological diversity in evolved swarms overall, which was beneficial in the most complex task environment.

Keywords

Quality Diversity

QD algorithm

Reference:

Collections

Masters

Full item page