This project is an attempt to understand and duplicate the echolocation and object recognition capabilities of bats by building an artificial sonar sensor system closely resembling the bats one and placing it on a mobile robot.
The sensor we propose here consists of one transmitter and two receivers. The pulse transmitted consists of a sinusoidal signal with a momentaneous frequency ranging from 40 kHz to 80 kHz. The signals as received by each ear will be processed with a bank of bandpass filters, the central frequencies of which span the frequency range of the transmit pulse. This setup mimics closely the first stages of the bat's sonar system. The outputs of this sensor system will be used to define monaural object localisation and object recognition based on the frequency dependence of the radiation pattern and the frequency dependent reflection characteristics, respectively.
Mounted on a mobile robot this sensor no longer produces isolated measurements but instead it delivers trains of measurements. By explicitly taking into account the sequential nature of the measurements additional constraints can be defined allowing biaural hearing along with all its advantages over monaural hearing. Preliminary results indicate that looking at trains of measurements will also make sophisticated object recognition as well as its extension to situation recognition possible. Finally, we intend to study systematically the different simplifications that result from behaviour bases control in processing the measurements from a sonar system mounted on a mobile robot as we suspect that bats make extensive use of similar strategies.