Creating an autonomous mobile robot with LEGO MINDSTORMS and leJOS which uses a genetic algorithm for path planning

Abstract

This research paper is a master thesis written at the University of Applied Sciences and Arts Northwestern Switzerland for the Master of Science in Business Information Systems course of studies. The master thesis mainly comprises a feasibility study in the area of autonomous mobile robotics combined with computational intelligence or more precisely evolutionary computing (i.e.: genetic algorithms). In this research paper, a profound literature research is performed and presented. Specific research questions are then formulated and addressed in an experiment. The path planning problem is one of the main research areas in autonomous mobile robotics. Algorithms such as Dijkstra’s algorithm or A* algorithm already solve the problem quite well under certain conditions. One of the biggest challenges is path planning in an unknown environment. Genetic algorithms can perform better than other algorithms when the environment is unknown. Since the experiment conducted for this research paper ran in an unknown environment, a genetic algorithm was developed to solve the path planning problem. The goal of the proposed algorithm was to show that the genetic algorithm not only performs better than simple trial and error, but that it also finds a suitable path in reasonable time. Besides providing a newly developed genetic algorithm for path planning, this research paper also analyses which aspects need to be considered when building an autonomous mobile robot which uses a genetic algorithm for path planning with only one proprioceptive sensor (odometry) and one exteroceptive sensor (ultrasonic). It also lists the advantages a genetic algorithm for path planning in combination with the chosen sensors provide in solving path planning problems simulated with a LEGO MINDSTORMS autonomous mobile robot. The main finding of the conducted experiment is that the combination of the robotic LEGO MINDSTORMS toolkit and the leJOS framework is not suitable for a sophisticated algorithm which requires precise sensor measurements. An ultrasonic sensor was initially deployed but quite often proved unable to detect obstacles. Subsequently, a touch sensor was used with the same algorithm and obstacles were again frequently not detected or, if they were, only after the obstacle had been hit several times. Due to this, the proposed genetic algorithm could not be tested sufficiently. Despite this, insights on which aspects to consider when building an autonomous mobile robot using a genetic algorithm for path planning with one proprioceptive sensor (odometry) and one exteroceptive sensor (ultrasonic) were gained.