Nearly Optimal Solution of HJB Equation Using Neural Networks: Applications to Control of DoD Systems and MEMS Assembly
Abstract
The goals of this grant were three. All have been accomplished. Goal 1 designed rigorous new nonlinear control schemes based on direct approximate solution of the Hamilton-Jacobi equations using neural networks (NN). On-line NN control techniques were developed that stabilize the system based on NN weight learning to approximate the optimal value function. Computational complexity was confronted using specialized structured NN controllers to provide efficient numerical solution algorithms for nonlinear optimal controllers. Optimal constrained controls were designed that satisfy actuator saturation limitations. Goal 2 proposed new information content and controllers for wireless networked systems. A new matrix-based discrete event controller was designed for wireless sensor networks with some mobile sentry nodes and some unattended ground sensors. The results were implemented on a mobile wireless sensor network testbed built at ARRI. Goal 3 built a prototype precision automated robotic microassembly system for future MEMS sensors and actuators for military networks. Novel control schemes and user interfaces were provided for tele-operated vision-guided microassembly.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 25, 2005
- Accession Number
- ADA436807
Entities
People
- F. L. Lewis
Organizations
- University of Texas at Arlington