Abstract:

An active sonar using Polaroid electrostatic transducers positioned at the end of a robot arm can recognize an object from its echoes. The sonar adaptively changes its location and configuration in response to the echoes. The sonar mimics biological echo-location systems, such as those employed by bats and dolphins, in that there is a center transmitter flanked by two adjustable receivers, the sonar has rotational and translational mobility, and the echo processing contains elements that have been observed in the mammalian auditory system. The sonar translates in a horizontal plane and rotates in pitch and yaw to position an object at a standard location within the beam patterns. The transmitter points at the object to maximize the incident acoustic intensity and the receivers rotate to maximize the echo amplitude and bandwidth and to minimize the echo-producing region. This procedure results in a unique echo vector associated with each object at a given object pose. Recognition is accomplished by extracting 32 values from the binaural echo patterns and searching a database that is constructed during a learning phase. The system can differentiate reliably the head and tail sides of a coin. [Work supported by NSF Grant IRI--9504079.]