Design and development of continuous-time recurrent neural networks evolution for cooperative distributed multi-agent systems

Sendra Arranz, Rafael (2021). Design and development of continuous-time recurrent neural networks evolution for cooperative distributed multi-agent systems. Thesis (Master thesis), E.T.S.I. Telecomunicación (UPM).

Description

Title: Design and development of continuous-time recurrent neural networks evolution for cooperative distributed multi-agent systems
Author/s:
  • Sendra Arranz, Rafael
Contributor/s:
  • Gutiérrez Martín, Álvaro
Item Type: Thesis (Master thesis)
Masters title: Teoría de la Señal y Comunicaciones
Date: 2021
Subjects:
Freetext Keywords: Swarm Robotics, Evolutionary Computation, Continuous-Time Recurrent Neural Networks, Recurrent Neural Networks, Minimal Communication, Emerged Communication, Genetic Algorithm, Natural Evolution Strategy, Neural Controller, Neuroevolution.
Faculty: E.T.S.I. Telecomunicación (UPM)
Department: Tecnología Fotónica y Bioingeniería
Creative Commons Licenses: Recognition - No derivative works - Non commercial

Full text

[img]
Preview
PDF - Requires a PDF viewer, such as GSview, Xpdf or Adobe Acrobat Reader
Download (6MB) | Preview

Abstract

Muti-agent systems are composed by multiple intelligent and distributed agents that interact in order to solve problems that would be utterly challenging individually. A type of multi-agent system that uses simple and locally interacting robots is swarm robotics, a class of collective robotics inspired by societies of insects. Swarm robotics is a field of research of constant growth and interest that combines concepts from artificial intelligence and robotics. It studies the use of many simple distributed robots that collectively cooperate in order to solve complex tasks. Moreover, the design of robust yet simple communication mechanisms, that allow the cooperation through direct interaction among robots, is an important aspect of swarm robotics systems. This Master Thesis explores the design and implementation of a minimal communication system, composed by a message and its underlying environmental context. To assess the performance and versatility of the communication, four benchmark swarm robotics tasks, that require communication at some extent, are designed and solved. The robot controllers, defining the behavior of agents, are based on Continuos-Time Recurrent Neural Networks (CTRNN) evolved using evolutionary computation algorithms. In particular, Genetic Algorithm (GA) and Separable Natural Evolution Strategies (SNES) are used and compared. All the experiments are carried out using a simulated robotics software designed and implemented within the frame of this Master Thesis. An important objective of this Master Thesis is the analysis of the communication that emerges as a result of the evolution process, in each experiment. It is shown that the swarm, whose robots are controlled by the evolved neural controllers, is capable of successfully solving the tasks of all the experiments. SNES outperforms GA in three of the four designed tasks, based on the obtained mean fitness scores. Apart from a detailed analysis of the emerged behaviors and communications, the scalability and robustness of the solutions are assessed in each experiment. The imposed tests expose that the evolved neural controllers fulfill both scalability and robustness properties, which are highly desired features of swarm robotics systems. Besides, the communication mechanics resulting from evolution are remarkably diverse. Specifically, it is shown that depending on the task, the communication can be purely situated, abstract or a combination of both.

More information

Item ID: 66353
DC Identifier: http://oa.upm.es/66353/
OAI Identifier: oai:oa.upm.es:66353
Deposited by: Biblioteca ETSI Telecomunicación
Deposited on: 12 Mar 2021 06:27
Last Modified: 12 Mar 2021 06:27
  • Logo InvestigaM (UPM)
  • Logo GEOUP4
  • Logo Open Access
  • Open Access
  • Logo Sherpa/Romeo
    Check whether the anglo-saxon journal in which you have published an article allows you to also publish it under open access.
  • Logo Dulcinea
    Check whether the spanish journal in which you have published an article allows you to also publish it under open access.
  • Logo de Recolecta
  • Logo del Observatorio I+D+i UPM
  • Logo de OpenCourseWare UPM