Date of Award
Spring 2026
Degree Type
Honors College Thesis
Academic Program
Ocean Engineering BS
Department
Ocean Science and Technology
First Advisor
Mustafa Kemal Cambazoglu
Advisor Department
Ocean Science and Technology
Abstract
Underwater robotics is a growing field with many practical applications, such as pipeline and offshore structure monitoring, deep sea mining, mine reconnaissance/removal, and marine environmental monitoring. USM’s Robotics Club will participate in the international RoboSub competition, where student-led teams build autonomous underwater vehicles (AUVs) that perform a variety of tasks within a pool requiring camera-based object detection. This project analyzes affordable hardware and software options for providing machine vision to USM’s future RoboSub AUV. The performance of an ESP32-CAM microcontroller, Raspberry Pi 5, and Jetson Orin Nano single-board computers running FOMO and YOLO object detection models was compared. These models were built on an AI machine learning platform called Edge Impulse, where hundreds of training images of each target, taken at varying distances and angles, were labeled and bounded. The setup for training and testing consisted of a camera looking through one side of an aquarium at targets on the opposite side. The ESP32 ran models through the Arduino interface with a small CSI camera, while the Jetson Orin Nano and Raspberry Pi ran models in their Linux terminals, with a USB camera. Comprehensive inference time and accuracy tests were conducted with model resolutions ranging from 32 to 320 for both FOMO and YOLO models. Performance estimations were conducted within Edge Impulse, with inference time estimates being relatively close to on-device performance, while estimated accuracy differed significantly from on-device performance. Physical testing, conducted at a variety of camera distances and angles from the target, demonstrated that ESP32 has lower than 40% overall accuracy and high inference times (up to 5924ms), suggesting it to be unreliable and
Copyright
Copyright for this thesis is owned by the author. It may be freely accessed by all users. However, any reuse or reproduction not covered by the exceptions of the Fair Use or Educational Use clauses of U.S. Copyright Law or without permission of the copyright holder may be a violation of federal law. Contact the administrator if you have additional questions.
Recommended Citation
King, Joseph, "Optimizing AUV Perception for Competitive Underwater Robotics: Analyzing Performance and Simplicity of Machine Vision Hardware" (2026). Honors Theses. 1141.
https://aquila.usm.edu/honors_theses/1141