Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


38218.pdf (7.49 MB)

ETD Abstract Container

Abstract Header

Modular Decentralized Genetic Fuzzy Control for Multi-UAV Slung Payloads

Bisig, Caleb R
http://orcid.org/0000-0002-5525-8602
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1617106491512366

Abstract Details

2021, MS, University of Cincinnati, Engineering and Applied Science: Aerospace Engineering.
Multi-drone slung payload systems that do not rely upon a leader for control instruction are a highly resilient option for critical delivery missions. In comparison to traditional control and modern deep-learning approaches, fuzzy systems are uniquely suited to balancing highly complex behaviors and high user readability post-training. By properly categorizing, training, and stacking purpose-driven fuzzy inference system (FIS) modules with separate cost functions or summed cost function components, unique behaviors can be developed and combined to form a powerful overall controller. In general, these have been referred to as genetic fuzzy trees (GFTs). Though technically the GFT proposed in this paper is a parallel set of four primary behaviors, the core program developed as a result of this research can and will easily add more tree-like series FIS modules to improve the responsiveness of the system to physical parameters. As a furthering of research in the field of genetic fuzzy decentralized control tasks, an early solution is explored for full three-dimensional point-mass control of a payload with neighbor avoidance safety and basic linear targeting behaviors. Drone teams of nominal counts (three to four drones depending on training task) are attached to a point mass payload by identical cables and are provided with visually obtainable state information. Isolated trainings build behaviors in tasks including cable pitch, altitude, cable yaw, and planar navigation, with a focus on visual representation of learned behaviors via fuzzy rule base surface plots. In response to difficulties experienced in setup of prior two-dimensional tests, a highly customizable object-oriented architecture dubbed GRAFT (Genetic ReArrangeable Fuzzy Tuner) for both simulation trial and fuzzy inference system creation has been developed and is explored briefly as a vital tool for future hand-built multi-inference system fuzzy controllers. This design exploits patterns in state space matrices, FIS definition and rule base requirements, and genetic algorithm genome sequence to create simulation trials and custom fuzzy trees with the ability to train, lock, add, or remove FIS modules from the GFT training structure at will. This auto-test generation feature permits custom individual or series of trials to be constructed from lists of possible initial states, creating intentional feature-space exploration approaches in quick succession rather than typical Monte-Carlo training. After the development of all behaviors with 3-4 drones, the system is then tested to examine performance in the face of drone power loss and resulting weight increase with teams of 5 and 7 drones. Some early tests of resiliency to sensor errors and noise are also analyzed. Lastly, recommendations for future fine-tuning and higher-level behavior implementation are discussed, with a plan for the next stage of research into bio-mimetic local minima avoidance using short-term shared position memory.
Ou Ma, Ph.D. (Committee Chair)
Kelly Cohen, Ph.D. (Committee Member)
Catharine McGhan, Ph.D. (Committee Member)
Anoop Sathyan, PhD (Committee Member)
103 p.
Aerospace Materials
Fuzzy Logic; Machine Learning; Genetic Algorithm; Drone; Decentralized; Slung Load

Recommended Citations

Citations

  • Bisig, C. R. (2021). Modular Decentralized Genetic Fuzzy Control for Multi-UAV Slung Payloads [Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1617106491512366

    APA Style (7th edition)

  • Bisig, Caleb. Modular Decentralized Genetic Fuzzy Control for Multi-UAV Slung Payloads. 2021. University of Cincinnati, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1617106491512366.

    MLA Style (8th edition)

  • Bisig, Caleb. "Modular Decentralized Genetic Fuzzy Control for Multi-UAV Slung Payloads." Master's thesis, University of Cincinnati, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1617106491512366

    Chicago Manual of Style (17th edition)

ucin1617106491512366
369
© 2021, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.
Release 3.2.12