PSO Tuner and Swarm Robotics Toolbox - Software Tools for Swarm Robotics Applications

Abstract

The Particle Swarm Optimization (PSO) algorithm is an stochastic optimization method that deploys a swarm of particles to explore and find the minimum of a cost function. In its most basic form, the algorithm can diverge depending on the parameters used. Two common solutions for this instability problem is the addition of an inertial constant and the constriction of the parameters through a group of equations that guarantee the convergence of the algorithm. An issue with these approaches is that they, in turn, depend on a group of parameters that need to be chosen carefully. To overcome that issue, we propose the use of a PSO Tuner, a specially trained recurrent neural network (RNN) that automatically sets the value of these parameters. In order to facilitate the design and use of the PSO Tuner, as well as to provide other useful tools for swarm robotics applications, we created a Swarm Robotics Toolbox. This consists of a set of functions, classes, scripts and interfaces that allow visualizing experimental results, saving figures, generating videos, performing statistical analyses, among others.