Convexification-Based Real-Time Robust Trajectory Optimization and Control for Aggressive Aerial Maneuvering
Abstract
The purpose of this action is to provide FY25 CR2 funds in the amount of $20K. GRANT#14014790.--The goal of this project is to build a mathematically rigorous convexification-based formulationand computational framework for reliable, real-time, robust trajectory optimization and controlfor aggressive autonomous aerial maneuvering in challenging mission scenarios with dynamicuncertainties and prohibitive resource constraints.Convexification refers to the process of transforming a non-convex trajectory optimization, i.e.,optimal control, problem into a single or recursively generated sequence of convex trajectoryoptimization problems. This process involves manipulating the original non-convex problemwhile both preserving its essential properties (e.g., ensuring the satisfaction of the originalproblem constraints) and ensuring convexity of the resulting optimization problem.Formulating a convex trajectory optimization has many theoretical and computational benefits.First and foremost, convexity ensures thatlocal optima are global optima, the fact of which hasled to the developmentof a solid theoretical foundation together with rigorousand efficientnumerical techniques for convex optimization problems. Further, convexification allows forsystematic handling of the uncertainties in the problem model, and hence enables thecomputation of optimal trajectories that are robust to model uncertainties by explicitlyaccounting for them. Once a convex optimization problem is formulated, it can be temporallydiscretized#i.e., converted from an infinite-dimensional representation to a finite-dimensionalrepresentation#extremely accurately and computationally efficiently. Then, we can utilize fastand reliable numerical methods for convex optimization, e.g., first-order projected gradientdescent (PGD) or second-order interior point method (IPM) algorithms, to obtain globallyoptimal solutions of the convex problems in polynomial time: There is an upper bound on thenumber of arithmetic operations to converge to the optimal solution for a given problem size.Thisdeterministic convergence behavior of convex optimization solution algorithms is the keyreason for their reliability and predictability. Consequently, building on these technical insights,we propose to develop the theoretical foundations and algorithmic capabilities to solve robusttrajectory optimization problems for aggressive aerial maneuvers in uncertain environments.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 10, 2025
- Source ID
- N000142512231
Entities
People
- Behçet Açıkmeşe
Organizations
- Office of Naval Research
- United States Navy
- University of Washington