One Tone, Two Ears, Three Dimensions: An investigation of qualitative echolocation strategies in synthetic bats and real robots
The aim of the work reported in this thesis is to investigate a methodology for studying perception by building and testing robotic models of animal sensory mechanisms. Much of Artificial Intelligence studies agent perception by exploring architectures for linking (often abstract) sensors and motors so as to give rise to particular behaviour. By contrast, this work proposes that perceptual investigations should begin with a characterisation of the underlying physical laws which govern the specific interaction of a sensor (or actuator) with its environment throughout the execution of a task. Moreover, it demonstrates that, through an understanding of task-physics, problems for which architectural solutions or explanations are often proposed may be solved more simply at the sensory interface - thereby minimising subsequent computation. This approach is applied to an investigation of the acoustical cues that may be exploited by several species of tone emitting insectivorous bats (species in the families Rhinolophidae and Hipposideridae) which localise prey using systematic pinnae scanning movements. From consideration of aspects of the sound filtering performed by the external and inner ear or these bats, three target localisation mechanisms are hypothesised and tested aboard a 6 degree-of-freedom, binaural, robotic echolocation system.In the first case, it is supposed that echolocators with narrow-band call structures use pinna movement to alter the directional sensitivity of their perceptual systems in the same whay that broad-band emitting bats rely on pinnae morphology to alter acoustic directionality at different frequencies.Scanning receivers also create dynamic cues - in the form of frequency and amplitude modulations - which very systematically with target angle. The second hypothesis investigated involves the extraction of timing cues from amplitude modulated echo envelopes.