A command computer for complex autonomous systems

Abstract

We describe a mode-commanding computer, CC, for autonomous complex systems that was inspired by the neuroanatomy and hypothesized functions of vertebrate brainstem reticular formations. A “mode” is a total action pattern of an organism, such as fight, flee, etc. CC is composed of K simple modules, 3 < K < 30, where each module is sparsely connected to other modules as well as to CC's sensory input systems. Each module's operation involves only local actions: squaring input probability values; averaging; normalizing the component sum of output vectors; and randomly respecifying the module's connections to other modules every computation cycle. CC's output consists of bundles of module outputs. Lengthy simulations have shown beyond doubt that CC always converges robustly and satisfactorily to one of M possible output mode indications in less than 30 computation cycles. M values in the range 2 < M < 18 have been used so far. Many future CC applications to autonomous vehicle control, robotics, etc., are expected, and an example is discussed.