Toan Vo

Summer Research Internship at the University of Utah

Tue, 20 Aug 2024 00:00:00 +0000

Introduction

This past summer, I had the incredible opportunity to work as an Undergraduate Research Assistant at the University of Utah’s School of Computing, in the Utah Artificial Intelligence and Robotics in Medicine (ARM) Lab led by Dr. Alan Kuntz. For three months, I dove deep into the world of medical robotics, focusing on advancing techniques in tendon-driven robotic systems. This experience not only honed my technical skills but also allowed me to collaborate with a brilliant team of researchers. My experience at the University of Utah was both academically challenging and immensely rewarding, helping me grow as a researcher and as a professional.

Predicting Tendon Robot Shape

Your browser does not support the video tag.

The data collection process of the tendon robot shown above

One of the main projects I tackled was developing a model to predict the shape of a tendon-driven robot. Under the leadership of Dr. Kuntz, the lab has been exploring these robots, which are fascinating because their shapes can be manipulated by adjusting tendon displacements, which are 4 strings wrapped around the tendon robot with different strengths, functioning much like muscles in a human body, pulling and shifting to create movement. However, accurately predicting these shapes, especially in a 3D space, is a complex problem that is ongoing at the time.

During my internship, I was tasked with developing a machine learning model that could accurately predict the shape of the robot based on tendon displacement data. The problem had been partially addressed using other techniques, but we needed a more robust and efficient solution, especially for real-time applications.

To address this challenge, I employed Long Short-Term Memory (LSTM) networks — a type of recurrent neural network that excels at understanding sequences. My goal was to create a model that could predict the robot’s shape as 3D point clouds based on the current tendon displacements. After rigorous testing and iteration, the model exceeded the previous learning-based techniques by 11% in terms of Chamfer distance, a metric used to measure the similarity between point clouds.

Deep Decoder Network (Left), the previous method, compared to LSTM (Right), where LSTM slightly performs better than the previous method

Collaborating on Bayesian Optimization for Surgical Retraction

Another exciting aspect of my internship was collaborating on a project that aimed to automate surgical retraction—a critical task in many medical procedures. Surgical retraction involves the use of tools to hold back tissue, providing the surgeon with better access to the area being operated on. Traditionally, this task requires manual input from surgical assistants, which can be imprecise and prone to human error. Our project sought to automate this process, using machine learning to optimize the positioning and force exerted by robotic retractors.

da Vinci Research Kit (dVRK) robot

The team I worked with adopted a Bayesian optimization approach to solve the problem, which is a powerful technique for finding the optimal set of parameters for a given system, particularly when the search space is large and complex. My role involved implementing different acquisition functions, which are used to determine the next point to sample in the optimization process. This approach led to significant improvements in detecting optimal attachment points for surgical tools, bringing us closer to fully automated surgical procedures.

Experiment of the surgical retraction on the dVRK simulation

By integrating various acquisition functions, we were able to find the function that improved the robot positioning and movement, allowing greater precision in revealing hidden areas of tissues more clearly.

Reflections

Reflecting on my time in Dr. Kuntz lab, I can confidently say that this internship was one of the most formative experiences of my academic career. I was fortunate enough to work with a group of talented researchers who constantly challenged me to push my boundaries and think critically about the problems we were tackling.

The work we accomplished—developing predictive models for tendon-driven robots and automating surgical retraction—has the potential to make a real difference in the field of medical robotics. While there is still much more work to be done, I’m proud of the contributions I made and excited about the future of this research. I left the internship with a renewed passion for robotics and machine learning, and a deeper understanding of how these fields intersect in the context of healthcare.

Overall, my summer at the University of Utah, under the mentorship of Dr. Alan Kuntz, was an unforgettable experience that has shaped my approach to research and problem-solving. I am incredibly grateful for the opportunity to contribute to such impactful projects and to work alongside some of the brightest minds in the field. As I continue my academic journey, I look forward to building on the skills and knowledge I gained during this internship, and I am excited to see where this research leads in the future.

Workspace Manager

Mon, 20 Nov 2023 00:00:00 +0000

In this project, I worked with a team to develop Workspace Manager, a tool designed to streamline workflow by allowing users to create and manage custom workspaces. The goal of the application was to help users quickly launch all the necessary applications, files, and tools for specific tasks with just one click, whether it’s for a coding session, a design project, or content creation. This project was built using JavaScript and MongoDB as the backend database.

The application supports multiple users, each with the ability to create personalized workspaces. With password-encrypted accounts, it ensures secure access and keeps your workspace configurations private. Users can easily add and store multiple file paths and applications within each workspace, so everything you need for a particular task is just a single action away.

Password encryption

Behind the scenes, Workspace Manager relies on MongoDB to handle data storage efficiently. When a user sets up their custom workspace, the application saves all file paths, application preferences, and user settings in the MongoDB database. This database structure allows for quick retrieval of information, making it easy for the application to launch the desired tools and files together when prompted.

The back-end logic, built with JavaScript, manages these interactions by communicating with the database to fetch the necessary data and execute commands to open the applications. This ensures a smooth and secure experience, as the password-encrypted accounts are also managed through MongoDB, safeguarding user configurations and workspaces. The front-end interface, developed with ClackJS, provides a seamless user experience, handling user inputs and delivering responses quickly by leveraging the data stored in MongoDB. Together, these components work in harmony to offer a streamlined and efficient workspace setup process.

Launching all workspace files at once

By consolidating the process of launching applications and files into a single action, Workspace Manager lets you dive right into your work, helping you maximize productivity and focus on what really matters. It’s an invaluable tool for anyone looking to enhance their workflow, reducing the setup time and allowing you to work more efficiently.

Experience

Tue, 24 Oct 2023 00:00:00 +0000

Impact of attendance on SAT scores

Sun, 25 Dec 2022 00:00:00 +0000

This analysis delves into the relationship between school attendance and SAT scores in New York high schools using data from 2010. The primary aim was to determine if higher attendance rates have a significant impact on students’ SAT performance and to explore potential differences between SAT reading and writing scores.

The study utilized two datasets: the SAT College Board 2010 results, which included mean scores for reading, writing, and math, and daily attendance records from the NYC Department of Education. By applying regression analysis, the research investigated the correlation between the average SAT scores and school attendance rates. Additionally, hypothesis testing was used to determine if there were statistically significant differences between the SAT reading and writing scores.

Our findings are as follows:

Impact of Attendance on SAT Scores: Our regression model revealed a positive relationship between school attendance and total SAT scores (sum of reading, writing, and math). The linear regression model predicted that a school with 100% attendance would have an average total SAT score of 1395.292. However, the residual plot suggested that the relationship may be non-linear, indicating that a quadratic model could better represent the data.

Difference Between SAT Reading and Writing Scores: Through hypothesis testing, we found a significant difference between mean SAT reading and writing scores (p-value < 2.2e-16). The mean difference was about 6.54 points, with reading scores typically higher than writing scores. This suggests that improving reading scores does not necessarily result in equivalent gains in writing scores.

Our analysis supported the idea that increased attendance has a positive impact on SAT performance, though further research is needed to explore the non-linear relationship indicated by the residuals. This project honed my skills in R programming, regression analysis, and hypothesis testing, while providing an opportunity to collaborate on meaningful educational research.

Highway crossing

Tue, 06 Apr 2021 00:00:00 +0000

Highway Crossing is an exciting 3D car game developed using C# and OpenGL, designed to challenge players as they navigate through busy highways filled with fast-moving traffic. The objective is simple yet engaging: guide your car safely across multiple lanes while avoiding collisions with other vehicles. As players progress through the game, they face increasingly difficult levels with more congested highways and faster-moving obstacles, requiring precise timing and strategic maneuvers.

The game features a range of 3D-rendered elements, including cars, roads, and various environmental objects, all created using OpenGL. The focus is on clear and functional visuals rather than hyper-realistic graphics, ensuring that players can easily identify obstacles and plan their movements. This design choice enhances the gameplay experience by providing a visually engaging world that supports the fast-paced nature of the game. To further immerse players, audio effects are integrated into the game, with sounds for engine revs and collisions adding to the overall atmosphere and making each drive feel more intense.

Texture

Highway Crossing also incorporates responsive controls and dynamic obstacles, ensuring that each playthrough offers a unique and challenging experience. Players must quickly adapt to changing traffic patterns and obstacles, testing their reflexes and strategic thinking. The game’s collision detection system ensures smooth interactions with the environment, making it a test of skill and concentration.

Throughout development, we collaborated closely, solving challenges such as optimizing object collisions, creating responsive controls, and synchronizing audio with gameplay. The project not only deepened my understanding of OpenGL and C#, but also honed my skills in teamwork, problem-solving, and game mechanics development.

Overall, Highway Crossing offers a blend of strategy, action, and quick reflexes within an engaging 3D environment. It demonstrates the potential of using C# and OpenGL to create interactive, enjoyable gaming experiences, providing a thrilling adventure for players as they race through congested highways.