Enhanced Visual Navigation for Mobile Robots

Abstract

Navigation solely by vision has been a goal in robotics for many years. It is desirable for a robot to plan and navigate the same way we do naturally as humans; however computation speed, changing illumination, complex algorithms, and a host of other problems have long plagued progress. One of the most prominent issues in using visible light outdoors and indoors is the unpredictable and changing ambient lighting. Shadows can vastly affect the robot’s ability to navigate on a simple path, and often look like separate objects to a robot. Features detected and described under perfect light may not register under dim light. This work focuses on a method developed to address these issues and construct a navigable path using color information, despite ambient and direct lighting. The results show a vast improvement over traditional intensity-based algorithms both in navigation and localization.