Abstract:

Navigation seems to automatically entail the construction and use of environment models. In this paper it is shown how, in certain situations, the information provided by an echolocation system contains invariants that can be used directly to control the navigation behavior of an agent without the need for an explicit environment model. In particular, the case of an echolocating agent moving through an opening between two reflecting objects is analyzed. The transmit signal, duration 2 ms, is a frequency modulated sinusoid, hyperbolic frequency sweep from 100 down to 20 kHz. The processing performed on the raw measurement results, i.e., the signals picked up by the microphones, is modeled on the peripheral auditory processing of the mammal brain. The output of the model is an approximation of the short-time amplitude spectrum of the received signal. It will be shown that the control of the agent can be expressed solely in terms of invariant properties of the patterns appearing on the outputs of this model. This approach not only predicts the flight path of bats flying between vertical wires but it also significantly simplifies the control of a mobile robot moving between obstacles as will be shown.