Cornelia Schulz and Andreas Zell
University of Tübingen, Germany
In robotics research, grid-based maps are widely used not only for localization and SLAM, but also as a prerequisite for navigational tasks such as path planning. Depending on the task, the utilized sensor, the chosen tradeoff between accuracy, runtime and memory requirements, and the exact map representation, a low grid resolution may be required. However, due to the involved discretization, occupancy grid maps suffer from aliasing effects and they are susceptible to sensor errors, especially if the resolution is low. In this work, we discuss different sub-pixel resolution methods, which improve the map quality without adapting the grid resolution. Therefore, we examine a grid traversal method inspired by antialiasing, which we use to apply weights to the grid cell updates in order to model straight beams more precisely. Further investigations include implicitly modeling the beam’s width and shape. Finally, we inspect methods developed for ONDT maps and conduct extensive experiments to evaluate the discussed techniques.