Autonomous mobile robot self-referencing with sensor windows and neural networks

Abstract

When navigating an environment a mobile robot can update its position and orientation by searching known landmarks and compare predictions with observations. This paper presents a method of mobile-robot self-referencing where every mapped object (obstacles to the global motion planner) in the environment can be used as potential sources of position and orientation information. This approach employs the efficiency of traversability vectors (t-vectors) for finding in-range geometric beacons and isolating surfaces visible to a sensor. Configuration-space (C-space) buffering (growing polygons to keep motion a safe distance from objects) will reduce the search time for finding in-range geometric beacons. Finally, a small multilayered neural network is used to provide a credence value for each predicted range that can be factored in to a filter or control strategy. This approach can be generalized to any ranging sensor that samples a region (e.g. IR sensors).