Optimal control-based mission simulation using accelerated automatic differentiation

Aerospace mission simulation and optimization present a variety of challenges that stem from their requirement to account for multiple facets. This thesis explores the use of modern, accelerated automatic differentiation in enhancing the solving and optimization of various aerospace problems posed as nonlinear optimal control problems. The growing complexity of aerospace systems and the limitations of traditional optimization workflows motivated the generalization of problem formulation, through the development of a JAX-based framework that efficiently implements and solves these optimal control problems through accelerated gradient-based optimization. After reviewing the theoretical foundations of optimal control and analyzing the role of gradients in optimization, the benefits of automatic differentiation are outlined, with the discussion of several state-of-the-art frameworks for the same. Through the modeling, discretization and optimization of 3 verifiable canonical problems - the Van der Pol oscillator, rolling disk, and free-flying robot - the direct transcription approach coupled with SciPy's Sequential Least Squares Quadratic Programming (SLSQP) solver is determined to be the most efficient problem setup. Then, JAX's automatic differentiation capabilities are implemented to enable fast and reliable computations of objective function and constraint gradients. A thorough comparison is drawn between the analytical and automatic gradient computations to cement the benefits of using JAX's capabilities and GPU-based optimization framework. The research is not centered around any one aircraft mission model or problem, but aims to generalize the application of the AD-based method. The future objective is to apply this to an RCAIDE mission framework that can simulate the operation of various types of aircraft. It is sought to prove that any problem can be optimized through the formulation of an objective function that covers all bases, with the implementation of JAX leading to huge cuts in computational time and overhead. Through the obtained results, we demonstrate that the JAX + SciPy SLSQP workflow achieves consistent convergence on problems ranging from simple to complex, offering a scalable and robust automatic differentiation framework for future aerospace mission optimization.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Aerospace Engineering

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Chilakapati, Akshaansh

Contributors dc:contributor

• Smart, Jordan

• Automatic Differentiation
• Optimal Control
• Trajectory Optimization
• Aerospace Simulation

Statement dc:rights

• Copyright 2025 Akshaansh Chilakapati

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/129349

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F129349