Brain-like intelligent control: from neural nets to larger-scale systems

Summary form only given. This paper summarizes progress in neurocontrol (neural networks for control), and provides a strategy to fill in the remaining gap between neurocontrol and large scale challenges such as the factory management challenge discussed by Albus and Meystel, with links to neuroscience. Neurocontrol has progressed in three main areas. First, we understand better how to improve stability both in neurocontrol and control in general-by using designs based on optimization over multiple time periods into the future. Second, there have been important applications, such as prototype ultraclean Ford cars driving in Detroit traffic, a controller which landed a huge MD-11 airplane with humans aboard and all control surfaces locked up, and the controller for the test flight of the US prototype of a hypersonic aircraft. Third, even more advanced “brain-like” designs have demonstrated superior performance in simulation and labscale benchmark tests. Much work will be needed to consolidate and extend these areas, and upgrade the subsystems in these designs. Nevertheless, even these “brain-like” designs lack certain capabilities called for by Albus and others, which are present in the mammalian brain. This paper proposes approaches to temporal chunking, spatial chunking and stochastic decision-making and imagination, to permit neural designs that fill in this gap