Employing an Embedded Renderer as Recognition Tool for Odometry, Map-Building, Navigation, and Localization on Active Sensing Robotics

Abstract

This study proposes a method that employs a renderer as a tool for environmental recognition. In the proposed system, features are extracted from sensors and cameras; the renderer represents scenes in a 3-D space to suit the purpose of the applications, and the applications resample the scenes to achieve their purpose after manipulating the renderer. As an example, this study presents implementation mechanisms of environmental recognition—odometry, map-building, navigation, and localization of automotive indoor robots. This method has a higher computational cost than typical feature-based methods; however, the algorithms are considerably intuitive. Although commercial rendering engines cannot be used as they are, a lightweight rendering engine dedicated to recognition can operate in embedded systems to enable real-time recognition. In addition, this study presents an experiment that corresponds to the simulation of moving robots indoors. In conclusion, this study proposes a change from the perspective of adopting a renderer–a well-established software technology that has been thoroughly investigated and can manipulate space–as an essential tool in the recognition framework.