More reliable second-order methods with applications to model predictive control
Abstract
Model predictive control (MPC) is a popular optimal control method used to guide rockets, missiles, and unmanned aerial vehicles. MPC problems are typically solved using second-order methods (optimization methods that use second-order derivatives to guide their search for an optimal solution). As applications of MPC at the DoD are often mission critical, it is important that they employ fast and reliable algorithms. Unfortunately, modern second-order methods are often to overfit benchmarks and can perform poorly on new instances. The aim of this research is to produce new second-order methods with stronger reliability guarantees that are also quicker than existing second-order methods. The effort will also build tools for tailoring these methods to MPC. To complement these two lines of research, investigations will be conducted as to how to best empirically evaluate theoretical predictions and algorithmic performance. This project is expected to result in significant advances in nonlinear optimization theory, new best practices for algorithmic development, and state-of-the-art open-source optimization software.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310242
Entities
People
- Oliver Hinder
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Pittsburgh