Undergraduate Research: Joshua Jowers and Another Approach to Drone Navigation- Spring Update

Headshot of Joshua

Author: Joshua Jowers | Major: Industrial Engineering | Semester: Spring 2023

My name is Joshua Jowers and I am an upcoming Senior majoring in Industrial Engineering with minors in data analytics and computer science. During the spring of 2023, I began undergraduate research with Dr. Chase Rainwater of the Industrial Engineering department. The overall goal of the research is to develop a method for autonomous drone navigation that is not reliant on GPS. During the spring, I began digging into code previously written for this problem by Wint Harvey during his time earning a PhD here at the University of Arkansas.

The motivation for this research is to develop an approach to drone navigation that is not susceptible to the weaknesses of GPS. While GPS is mostly effective and is currently the most used approach for autonomous drone navigation, it has a couple of major vulnerabilities: signal blocking and spoofing. GPS signals can be disrupted naturally by objects such as buildings, mountains, and trees, or artificially by an enemy with malicious intent. The latter situation is prevalent in military settings which provides strong motivation for a navigation method that doesn’t use signals that can be intercepted.

This is where machine learning comes in. Wint Harvey, working with Dr. Rainwater, created a convolutional neural network (CNN) using python code and a library called TensorFlow that can be trained on images of a given area and then make predictions of the specific location of a new image from that area with.

During the spring of 2022, I participated in a course designed to help Honors Industrial Engineering students pick a research mentor and topic. When Dr. Rainwater presented some of the projects that he has been involved in during this class, I was very interested in this location prediction project, so I followed up with him on it which led to me joining on this project.

When I began looking into the code, I quickly realized that there was a lot for me to learn. I only had a little bit of experience with Python and none with machine learning. Jumping straight into this project has been a lot to handle as I am dealing with hundreds of lines of code and massive datasets while trying to learn it all.

The original plan for the project was to quickly get me familiar with the TensorFlow code and begin work on expanding Wint’s code to take in a series of photographs as if they were taken along a path and to make predictions that consider the previous predictions. However, after several weeks of meetings with Dr. Rainwater and many hours of debugging, we shifted the first objective to porting the TensorFlow code over to a different machine learning library called Pytorch.

This decision was made primarily because Pytorch has more features and documentation relating the specific code I am working on. Most of the spring semester was spent learning Pytorch and rewriting the code to work using this new library. Once this process is finished, I will refocus my efforts to adding the image sets functionality in the fall semester. This will be accomplished by using a specific machine learning capability called Long Short-Term Memory (LSTM). This means that each time that the machine learning model makes a prediction of a new image, it will take into consideration the previously predicted locations and the amount of displacement due to the drone’s movement.

Overall, the Honors College Research Grant has allowed me to spend time digging deep into understanding this problem and machine learning as a whole. While there have been more challenges that I anticipated, it has been really rewarding getting to work through those with the support of Dr. Rainwater. I will continue to work on this code and this problem through the fall semester, and I look forward to learning more and seeing the results of my effort thus far.