A grid-based approach to localization for robotics applications

From robotic probes on Mars to self driving cars here on Earth, the process of moving around the planet begins with localization. Before an autonomous vehicle can decide where to go, it must first determine where it is and what obstacles may be nearby. By using various sensors, an autonomous platform can gather the data it needs to identify its position and orientation within its surroundings. This position and orientation is known as the pose of the robot. In this report, we will review a particle filter based algorithm for localizing an autonomous mining robot in a known 2D map. This algorithm uses input from a LIDAR sensor that provides range information for every 0.25—¦ in a 180—¦ arc. We will also discuss how this algorithm\u27s pose outputs were used to guide the autonomous operations of the Mountaineer Mining Vehicle, WVU\u27s 2014 winning entry to NASA\u27s Robotic Mining Competition. Sufficiently fast runtimes were achieved with small deviations from the correct pose. These runtimes were further improved by culling bad particles after a partial evaluation of the sensor data. This resulted in a 30-40% reduction in runtime without increasing the error