Files and links (1)
CC BY V4.0, Open Access
Thesis
NEURAL NETWORK APPROXIMATION OF STEADY-STATE AWARE MODEL PREDICTIVE CONTROL FOR DRONE NAVIGATION IN RESOURCE-CONSTRAINED ENVIRONMENTS
Washington State University
Master of Science (MS), Washington State University
12/2024
DOI:
https://doi.org/10.7273/000007181
Abstract
This project focuses on developing a neural network-based approximation for Steady-State-Aware Model Predictive Control (MPC), specifically designed to enhance real-time control performance for quadcopters. By ensuring that the system converges to the desired configuration if it is admissible or to the best admissible configuration, steady-state-aware MPC ensures output tracking and constraint satisfaction in all situations.
To further reduce the real-time computational complexity of solving optimization problems that MPC poses for systems with limited computing power, a neural network was trained to approximate the control law of steady-state-aware MPC. This network was trained using data from a steady-state-aware MPC algorithm, which, once trained, could predict control actions based on current and reference states in real time, replicating the behavior of the MPC without the associated computational burden. The neural network-based controller was integrated into the quadcopter’s control loop and tested in both simulation and real-world environments. Results from simulation and experiment demonstrated that the neural network maintains stability provided by steady-state-aware MPC with shorter computation time.
Metrics
2 File views/ downloads
3 Record Views
Details
- Title
- NEURAL NETWORK APPROXIMATION OF STEADY-STATE AWARE MODEL PREDICTIVE CONTROL FOR DRONE NAVIGATION IN RESOURCE-CONSTRAINED ENVIRONMENTS
- Creators
- Joshua Folorunso
- Contributors
- Mehdi Hosseinzadeh (Chair)John Swensen (Committee Member)Ming Luo (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- Voiland College of Engineering and Architecture
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University
- Number of pages
- 72
- Identifiers
- 99901195201301842
- Language
- English
- Resource Type
- Thesis