Constructing neural networks for contact tracking

Abstract

A neural network approach for contact state estimation is presented. This neural network, NICE (neurally inspired contact estimation), has been constructed to directly embody the major problem domain constraint of uniform contact velocity and heading. NICE networks are constructed, not trained, to estimate contact position and motion from angle-of-arrival (AOA) measurements. The major advantages of the NICE system over existing methods are execution speed, an assessment of solution sensitivity, and the potential for sensor fusion. This system offers a number of attractive features. Foremost, a bearing line constrains the locus of points where a contact might be at a given time. Furthermore, different AOA sensors merely produce different loci; all are equivalent and can be fused using the system. Intermittent data can be accommodated by configuring correlation neurons to ignore the missing data, and the geographical grid resolutions can be varied to adjust to the quality of the sensor readings. In addition, the neural network can be executed in a highly parallel manner, taking advantage of the state-of-the-art parallel hardware.<>