Fast computational processing for mobile robots’ self-localization Hélder Ribeiro, Pedro Silva, Ricardo Roriz, Tiago Maia, Rui Saraiva, Gil Lopes and A.Fernando Ribeiro Laboratório de Automação e Robótica, Grupo de Controlo Automação e Robótica Department of Industrial Electronics, University of Minho Azurém, Guimarães, Portugal { a58795, a68541, a68536, a57126, a58783}@alunos.uminho.pt, {gil, fernando}@dei.uminho.pt Abstract—This paper intends to present a different approach to solve the Self-Localization problem regarding a RoboCup’s Middle Size League game, developed by MINHO team researchers. The method uses white field markings as key points, to compute the position with least error, creating an error-based graphic where the minimum corresponds to the real position, that are computed by comparing the key (line) points with a precomputed set of values for each position. This approach allows a very fast local and global localization calculation, allowing the global localization to be used more often, while driving the estimate to its real value. Differently from the majority of other teams in this league, it was important to come up with a new and improved method to solve the traditional slow Self-Localization problem. Keywords—RoboCup; MSL; Middle Size League; MINHO team; Self-Localization; Localization; I. INTRODUCTION MINHO team started a robotic football team in 1997 and has been participating on RoboCup scientific challenge since 1999 making improvements on their platforms and software. In 2011 that development paused and returned in 2014. The restart consisted in rebuilding both hardware and software, and there was an urgent need to improve the robots self- localization technique, and to push the development of a new method. Regarding the RoboCup MSL specific application autonomous robots need to know their position on the field (the world), to be able to move to a certain position, and to kick towards a certain direction or to perform high level agent coordination, but that implies the need of the existence of a method that allow the robot to self-localize, only using local on-board sensors. The robustness, processing time and false positives in the line point detection are a big concern, and all of them model the structure and procedure of the method. At the moment the majority of teams use the self-localization algorithm created by Brainstormers Tribots [1]. The method here presented tries to improve the computational time using a rather different approach in the error calculation procedure. The method uses a precomputed set of meaningful distances of every possible position in the 20x14m field area (standard 18x12 plus 1 meter all around the field, resulting in a 20x14m world), with a 10x10cm resolution. Then, given the robot’s true orientation, the position with the least error is computed using an error modelling function, coming up with the true position of the robot on the field. In Section II, a tracking of the league’s development is given, while in Section III the imaging solution (also a standard in the league) and the method used in the search for interest points in the image, is presented. Section IV presents the method itself, explaining the world view from a certain point in the field, and the error calculation procedure. Section V explains the two different methods for acquiring the robot’s true heading, in relation to a known reference, one by hardware (with its complementary software) and the other by software. Section VI addresses the results achieved during the research and application of the method, concluding this work. II. LEAGUE’S DEVELOPMENT The league evolved rapidly throughout the years, accomplishing new challenges and tasks, complying with the current advances of modern day computing technologies. With new camera technologies, new computers and better communications, the imaging quality, the processing power and the information sharing velocity has been greatly improved. Due to these facts, the league stepped up new challenges to the teams, making changes in the rules and in the composition of the field. As the global system is faster, the necessity of having robust, effective and fast algorithms (to accomplish different tasks) to fit the “processing time window” is urging. Regarding the field layout, it is now larger, with 18x12 meters playable area, only with the standard white line markings. Taking into account that the robot’s catadioptric camera system only covers about 4 meters radius of field area, there is a lot of information missing, giving a lot more importance to team communication and agent coordination, to accomplish team and tactical objectives of the game. Stated that the league, and the whole game, is evolving fast, the development of the present method, represents the evolution of a well-settled algorithm, to improve, at least, the computational time involved in the self-localization process. III. IMAGING SOLUTION AND POINTS OF INTEREST A. Catadioptric Sensor Setup When it comes to imaging, there is a standard in the league, to use a catadioptric sensor, which is obtained by pairing a catadioptric mirror and a camera, with various technologies, output types and price ranges. The catadioptric mirror [2] [3] was developed specifically for this application, using simulation tools to achieve best performances. The image provided by this imaging setup arrives at a frequency of 30Hz, with a cropped resolution of 480x480, in YUV 411 format. Using camera’s setup tool, parameters can