Extended Particle Swarm Optimization by Partial Randomization for a Multi-Robot System

Abstract

This paper investigates a particle swarm optimization (PSO) approach to perform a cooperative task in a multirobot system. PSO is known to be relatively easy to implement and efficient in solving various optimization problems. However, it has also been reported that the PSO approach is inefficient in solving large-scale multimodal problems. To overcome this unwanted characteristic of PSO, we have been developing a novel method that performs partial initialization in a small probability and demonstrating its basic effectiveness for benchmark functions. This paper further discusses a real world problem. We tackle the control of autonomous mobile robots by PSO. Similar to typical methods in evolutionary robotics, artificial neural networks (ANNs) are employed as controllers of each robot. PSO is utilized as a methodology for designing the ANNs. To examine our approach, computer simulations are executed using a cooperative box-pushing task. The results obtained are compared with standard PSO results. It is found that our proposed approach yields competitive results.