An active sensing method using estimated errors for multisensor fusion systems

An active sensing method for multisensor fusion systems with actuators is proposed. To realize active sensing with multiple sensors, i) where to position sensors, ii) how to associate data, and iii) how to fuse data should be determined. The authors propose a new method mainly concerning i). The method utilizes estimated errors of estimated values to determine optimal sensor locations where useful data are expected to be obtained and effectively associated. An algorithm to calculate nearly optimal sensor locations, instead of exact optimal locations, is also proposed to reduce calculation. As examples, the active sensing method is applied to multi-target tracking by a system with two hand-eye cameras, and visual and tactile fusion in a system with a camera and a tactile sensor. By using this method, the sensing strategy is optimized for the object of measurement.