A distributed architecture for active perception in autonomous robots

Abstract

Describes the developed architecture for sensor information integration in a mobile robot. Hardware and software modules work locally with sensors signals, distributing the fitness functions over a parallel processing scheme. This organization tries to overcome real time constraints when dealing a with dynamic system-environments. A distributed network of DS5000 Dallas microprocessors and T800 transputers assisted by several DSP's, is addressed for first stages of sensors signals accommodation and generation of low level survival behaviors. Human knowledge in transferred to the perception modules as fuzzy logic inferences located on each processor. A set of decision rules account for efficient decisions making using a small number of linguistic terms that condense rough sensor data by means of membership functions representation. System experience can be modulated with existing algorithms according with some pre-defined safety conditions.