We have developed a 5 DOF ultrasonic sensorhead that allows the study of different sensor-motor loops in echolocation. The head consists of a neck, allowing for the panning of the head as a whole, a transmitter which is fixed to the head and two independently orientable receivers, one on either side of the transmitter. Each receiver has two DOF, allowing for pan and tilt movements around its centre. The generation of the transmit signal, the processing of the received signals as well as the motorcontrol of the sensorhead is executed on a transputer network. To process the received signals we make use of a filterbank modelled on the bat's cochlea. This filterbank consists of a number of bandpass filters logarithmically distributed over the frequency range of interest, currently from 30 kHz to 100 kHz. The shape of these filters and the distribution of their central frequencies can be varied depending on the specific echolocation task being investigated. The instantaneous amplitudes of the outputs of these filters is determined by a rectification and low-pass filtering process. We believe that although this is only a crude approximation to the actual processing performed at the cochlear level it nevertheless succeeds in modelling the cochlea's most important characteristics. The outputs of this model are the inputs for the different echolocation tasks we want to study. The head is now being used to investigate what clues are available to a binaural animal while tracking a prey. We are also looking at the robustness of the different tracking schemes in the presence of clutter. It is clear that the head allows the study of many other mechanisms, especially those involving sensor-motor interaction, important to echolocation. Hence we argue that bionic device such as our sensorhead are valuable tools adding to the store of instruments for the study of echolocation by biological organisms.
Abstract: