The capstone project we have been working on these past two semesters is to create a Nonlinear Sprayer Pump Control Algorithm that optimizes pressure and flow to reach a target speed. The company we work for is John Deere and Peter Martinson, our company liaison who helps guide us through problems and is our main point of contact with John Deere. We are implementing a Pfleider-Peterman model in conjunction with the Pump Model to control the inputs and outputs of the pump.

Rohde & Schwarz proposed a project that would explore utilizing machine learning to facilitate the communications industry's move toward 6G. Digital Pre-Distortion (DPD) is a widely used technique to enhance the linearity of power amplifiers in communication systems. Our team developed a machine learning-based DPD algorithm that maintains signal quality (measured by Error Vector Magnitude (EVM)), is generalizable across various systems, and offers faster performance compared to iterative DPD algorithms by experimenting with different RNN architectures.

Our project is to design an e-bike controller which interfaces with existing components of an e-bike kit. This included the design and implementation of the motor drive system, accompanying circuitry and its control and peripheral code which interfaces with the motor and accessories of an e-bike kit. Our bike features a throttle, display interface from which the user can view important ride statistics, information about the overall system health, current location, and buttons with which to navigate through the user interface. 

Team Members

6G networking is steps away! Joint Communication and Sensing v2.0 has continued the work of creating a testable and scalable system to optimize and refine communication and sensing processes. This involves optimizing both a testbed system and eventually establishing a stronger foundation for managing frequency better. Our project builds off of an existing testbed system that utilizes a CPU-run algorithm and a radio to gather, analyze, and identify usable frequency. The intent of v2.0 was to expand the testbed and make improvements to such processes.

The Photonic Biosensor Project (PBP) expands on a functional desktop-based system that accurately detects SARS-CoV-2 SP and influenza NP. We set out to make the system portable (broadening its usability), reusable (lowering operating costs), and extending its application into depression detection (greater functionality). To make the system portable, we integrated a variety of internal components to a central computer and made it all controllable via an intuitive UI usable by any lab technician.

While unit testing is widely used to test source code quality, inline testing introduces a new granularity of testing software more suited to the level of individual program statements. For our project, we developed a decision tree machine learning model to search for program statements well-suited to I-Test, the first inline testing framework developed by Yuki Liu, Pengyu Nie, Owolabi Legunsen, and Milos Gligoric.

Re:Form seeks to provide valuable and informational feedback to trainers and experienced athletes on an individual’s form during a particular movement. With the ability to provide information that is difficult to gauge from sight alone, Re:Form integrates data from an array of sensors and displays feedback on an iPhone application that allows the user to evaluate their form, crucial for further form improvements and injury risk reductions.

Arome is designed to be an automatic dietary monitoring system that uses a sensor array consisting of multiple Metal Oxide (MOX) chemical sensors. These sensors can detect gasses named Volatile Organic Compounds (VOCs) such as acetone and toluene, and have been successfully used to quantify blood glucose, diagnose pulmonary disease and detect asthma from human breath. The system will use the sensor array to identify different foods based on their emitted VOCs when broken down and display the results on a mobile application.

Team Members

New environments require mapping for better understanding by the agents exploring them. In this project, we carry out environment exploration with a few agents operating on edge devices (NVIDIA Jetson Nanos) that perform Federated Learning with state-of-the-art Depth Estimation models. For training we collect real-time RGB images and depth information during exploration of various rooms around the UT Campus using RGB-D cameras. We then use Federated Learning and Machine Learning models to create Depth Estimation images based off the RGB images we collect of new rooms. 

Heat stroke and dehydration are critical health situations which many people may not identify until their condition is serious and requires a hospital visit. These conditions disproportionately affect the elderly, young children, and those with poor health. For this project, we developed a network of wearable sensors that monitor an individual’s heart rate, body temperature, galvanic skin response, and ambient temperature and humidity to determine their risk of heat stroke and dehydration.