Navigation By Correlating Geometric Sensor Data

A continuing problem in mobile robotics is that of achieving reliable autonomous navigation based only on information obtained from the sensors of a mobile vehicle. The basic navigation problem based on the observation of navigation beacons has been studied extensively over many hundreds of years, and is in general well-understood. The application of these techniques in robotics has faltered on the problem of reliably extracting beacons from sensor data and utilizing them in automating the navigation process. In this paper we offer a solution to the mobile robot navigation problem, which relies on the concept of a ugeneralized geometric beaconn - a feature which can be reliably observed in successive sensor measurements (a beacon), and which can be described in terms of some small number of geometric objects. This navigation algorithm is based around a simple Kalman-filter which is employed to maintain a map of these observed geometric beacons, and into which new sensor measurements can be matched. We describe three different implementations of this navigation algorithm, the first on a vehicle with only one rotating sonar, the second on a vehicle with six static sonars, and the third on a vehicle equipped with both a sonar and an active infra-red sensor. These implementations demonstrate how different geometric beacons extracted from different sensors and algorithms can be used to provide a robust and reliable estimate of mobile robot location.