Neural Networks for Control of Autonomous and Semi-Autonomous Systems

点击翻译按钮获取中文摘要

Under this grant, the PI will try to develop fundamental, qualitative extension, and the advanced adaptive dynamic programming (ADP) designs he has applied very effectively to aerospace problems in the part. He will focus on four major areas:1) multilayer neural network based stochastic optimal controllers with bounds on the error of the cost function 2) simplified adaptive critic based controllers for nonlinear systems3) new observer/filters for nonlinear systems with a unified formulation and 4) new bio-inspired problem solving structures for coupled (dynamically or task wise) systems.He will attempt to apply these methods to the control of connected and complex systems, and based on mathematically rigorous solution and analysis structures. The test problems involve elements of 'robust and yet fragile' complex systems in an autonomous and semi-autonomous setting. The outcome of this research will help understand the properties of this important class of problems, some examples of which are the telecommunications, power grids, civil infrastructure, supply chain management, internet, biological systems etc. It should be noted that the traditional approaches, in use for a long time, have not provided the key to understanding or complete solutions of this classes of nonlinear problems with uncertainties in dynamics and goals. The educational part of this research will deal with relating this outcome to K-12 students through interactions with local science and math teachers. With the experiments that are proposed in this research, the K-12 students might be able to relate better to the use of math and science in Engineering.