New Adaptive Dynamic Programming Structures From Neurocognitive Psychology and Graphical Games

University Of Texas At Arlington, Arlington TX

Investigators

Abstract

This project brings together a top control engineer with a cognitive neuroscientist, in order to design a new family of control designs which would replicate and explain key capabilities of living brains which have never yet been achieved in engineering (or in the models used in computational neuroscience). This new work builds on previous work by Frank Lewis, in developing adaptive controllers (RLADP) which can maximize performance over time, in the face of nonlinearity and challenges which require foresight, such as the management of power grids. Here, the group will try to explain and replicate how brains can also handle challenges which require them to learn how to structure time, with multiple levels of decision with multiple time horizons, and how to handle complex structure in space, as we need to in managing complex infrastructure networks. These two challenges essentially address two of the three gaps between today's best RLADP and the highest capabilities of the mammal brain. The new fundamental design work will feed into ongoing laboratory work in the control of electric power microgrids, which are important as building blocks for future electric power distribution networks capable of coping with large penetrations of plug-in hybrid cars or rooftop solar and the like. The previous work is reviewed in the Handbook of RLADP, from IEEE/Wiley, edited by Lewis and Liu. Spatial complexity will be addressed by considering optimal control of systems defined over graphs, such as power and communication networks. Game theoretic extensions, related to issues of distributed or collective intelligence, will also be considered. The work will also build on work by co-PI Levine on models of mechanisms in the brain involving the emotional gates in the amygdala and deliberative decisions in the anterior cingulate cortex.