The overall objective of this study is to provide online robust adaptive dynamic programming (ADP) based optimal controllers with guaranteed performance, supported by a rigorous design and mathematical framework, and without utilizing policy and value iterations, for unknown linear and nonlinear networked control systems (NCS). The approach taken here employs adaptive network learning as a fundamental block and utilizes past history of cost-to-go information, and updates the control input once a sampling interval in a forward-in-time manner without using a system model and offline learning phase for the NNs.
Intellectual Merit The proposed research presents an opportunity to deal with a more powerful and unified paradigm of complex learning problems and envisions a brain-like controller. The proposed effort will advance the state of the art in ADP for control and guarantees stability and performance in the presence of not only uncertain system dynamics and disturbances, but also network imperfections such as random delays, packet losses and quantization errors without using iterative approach.
Broader Impact This effort would directly impact all real-time practical systems such as the efficient operation and energy security of the smart grid, near zero-emission automotive control systems, and next generation manufacturing system. Such control schemes are required for global competitiveness of the US industry. Technology transfer will occur through the NSF Industry/University Cooperative Research Center on Intelligent Maintenance Systems. Within the research community, this work will inspire more theoretical results while providing training opportunities to next generation students, future scientists and engineers including from underrepresented groups.