Performance evaluation and optimization of swarms of robots in a specific task

Márquez Vega, Luis Ángel (2019) Performance evaluation and optimization of swarms of robots in a specific task. Maestría thesis, Universidad Autónoma de Nuevo León.

Vista previa

Texto 1080313958.pdf - Versión Aceptada Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (21MB) | Vista previa

Resumen

Objectives and methodology: Nowadays the swarms of robots represent an alternative to solve a wide range of tasks as search, aggregation, predatorprey, foraging, etc. However, determining how well the task is resolved is an important current problem, assign evaluation metrics to tasks performed by swarms of robots is very useful in order to measure the performance of a particular swarm in the task resolution. Find the control parameters of a swarm of robots that resolves a task with the best possible performance represents many benefits as saving of energetic resources and time. The general objective in this thesis is to evaluate and improve the performance of a swarm of robots in the resolution of a particular task, for that reason the following specific objectives are proposed: 1) To describe a flocking task with target zone search and to determine evaluation metrics that measure the task resolution; 2) To implement behavior policies for a simulated swarm of quadrotors; 3) To implement multi-objective optimization techniques in order to find the best sets of control parameters of the swarm that resolve the proposed task with the best possible performance; 4) To compare the performance of the implemented multi-objective optimization algorithms in order to determine which algorithm represents the best option to optimize this type of tasks. Different methods to control swarms of robots have been proposed, in this thesis a bio-inspired model based in repulsion (∆r), orientation (∆o) and attraction (∆a) tendencies between biological species as bird flocks and schools of fish is applied in the simulated swarm of quadrotors. Different experiments are proposed, the flocking task with target zone search is optimized for swarms of quadrotors of 5, 10 and 20 members and with two different conditions in the environment, one case without obstacles and another case with obstacles in the arena. The task is evaluated by four proposed objective functions formulated as minimization problems which are oriented to reach four main objectives in the task, as these objectives functions are minimized the desired behavior of the swarm of quadrotors is reached. The Multi-Objective Particle Swarm Optimization (MOPSO), the Nondominated Sorting Genetic Algorithm II using Differential Evolution (NSGA-II-DE) and the Multiobjective Evolutionary Algorithm based on Decomposition using Differential Evolution (MOEA/D-DE) are used to optimize the control parameters ∆r, ∆o and ∆a for the proposed task in each experiment. The Hypervolume measure (HV ), a modified C-metric (Q) and the time per cycle (T P C) are the selected metrics to evaluate the performance of the multi-objective optimization algorithms. Contributions and conclusions: The obtained results show that the selected behavior policies produces collaborative interactions between members of the swarm that benefit the resolution of the task. Use multi-objective optimization techniques directly on the quadrotor swarm simulator produces small number of optimized solutions because the optimization process is only suitable with small populations and with a reduced number of cycles due to the...

Actions (login required)

Ver elemento