When navigating an environment a mobile robot can update itsposition and orientation by searching known landmarks and comparepredictions with observations. This paper presents a method ofmobile-robot self-referencing where every mapped object (obstacles tothe global motion planner) in the environment can be used as potentialsources of position and orientation information. This approach employsthe efficiency of traversability vectors (t-vectors) for findingin-range geometric beacons and isolating surfaces visible to a sensor.Configuration-space (C-space) buffering (growing polygons to keep motiona safe distance from objects) will reduce the search time for findingin-range geometric beacons. Finally, a small multilayered neural networkis used to provide a credence value for each predicted range that can befactored in to a filter or control strategy. This approach can begeneralized to any ranging sensor that samples a region (e.g. IRsensors)
展开▼
