Feature-Based Localization in Sonar-Equipped Autonomous Mobile Robots Through Hough Transform and Unsupervised Learning Network

摘要

As we approach the new millennium, robots are playing an increasingly important role in our everyday lives. Robotics has evolved in industrial and military applications, and unmanned space exploration promises the continued development of ever-more-complex robots. Over the past few decades, research has focused on the development of autonomous mobile robots - robots that can move about without human supervision. This brings with it several problems, however, specifically the problem of localization. How can the robot determine its own position and orientation relative to the environment around it. Various methods of localization in mobile robots have been explored. Most of these methods, however, assume some a priori knowledge of the environment, or that the robot will have access to navigation beacons or Global Positioning Satellites. In this thesis, the foundations for feature-based localization are explored. An algorithm involving the Rough transform of range data and a neural network is developed, which enables the robot to find an unspecified number of wall-like features in its vicinity and determine the range and orientation of these walls relative to itself. Computation times are shown to be quite reasonable, and the algorithm is applied in both simulated and real-world indoor environments.