Research Projects

Research Experience for Undergraduates (REU) Research Projects

The proposed research projects are subject to change, contingent on new mentors and/or the research interests of the students.

Learn about past research projects in the Research Projects Archive.

Center for Autonomy
Autonomous Systems Group

Pathfinding and RL in PacMan. Pathfinding is an essential problem in artificial intelligence, with applications spanning robotics, logistics, and interactive environments. The project begins with developing and implementing classic search algorithms such as Breadth-First Search (BFS), Depth-First Search (DFS), and A*. Students will then transition to reinforcement learning (RL), exploring algorithms like Q-learning to enable agents to learn optimal behaviors through interaction with their environment.

Prerequisites:

  • Mathematics: Probability, linear algebra, and basic calculus (useful for understanding RL concepts)
  • Programming: experience with Python

Learning outcomes:

  • Gain basic understanding of Markov Decision Processes (MDPs) and dynamic programming
  • Implement and analyze classical search algorithms for pathfinding
  • Learn the fundamentals of reinforcement learning, including value iteration and policy learning
  • Implement Q-learning and evaluate agent performance in the Pac-Man environment

Mentor(s):

  • Caleb Probine, Graduate Student
  • Quentin Rommel, Graduate Student

Center for Autonomy
Autonomous Systems Group

Self-Driving Car: From Cruise Control to Obstacle Avoidance. How does a self-driving car decide when to brake, how hard to steer, and how to avoid obstacles, all in real time? This project explores these questions by building controllers for a simulated car, starting simple and adding complexity step by step. Students will first implement a controller that keeps the car at a constant speed, then one that follows a curved road. As the scenarios get harder (speed limits, steering limits, obstacles on the road, and more realistic car physics) students will discover why simple controllers break down and build more powerful ones to handle each new challenge.

Prerequisites:

  • Mathematics: Basic linear algebra, basic calculus
  • Programming: Experience with Python or MATLAB

Learning outcomes:

  • Understand how to model a car's motion mathematically
  • Implement and tune controllers for trajectory tracking
  • Learn why constraints and nonlinear dynamics require more advanced control methods
  • Implement model predictive control to handle obstacles and physical limits
  • Visualize and compare controller performance across increasingly complex driving scenarios

Mentor(s):

  • Filippos Fotiadis, Research Associate

Center for Autonomy
GeoElements

Real-Time Digital Twin Framework Using Gaussian Splatting and Graph Network Surrogates for Dynamic Simulation. Real-Time Interactive Digital Twins Using Gaussian Splatting and Graph Neural Networks. This research project develops a novel framework for creating interactive digital twins from video input. By combining 3D Gaussian Splatting with Graph Neural Networks (GNNs), we aim to build a system that can reconstruct and simulate real-world environments in real-time.

Our pipeline processes video streams through three key stages:

  1. Object segmentation from video frames to identify distinct entities
  2. 3D reconstruction using Gaussian Splatting for efficient scene representation
  3. Physics simulation via GNN-based surrogate models for real-time interaction

The innovation lies in integrating geometry-aware 3D reconstruction with learned physics simulation, enabling rapid updates between the physical environment and its digital twin. By using GNNs as ML surrogates for traditional physics engines, we can achieve sub-second latency while maintaining accuracy.

The project emphasizes practical applications in robotics, augmented reality, and smart infrastructure. Our implementation will leverage modern tools including PyTorch for deep learning, Open3D for 3D vision, and game engines for visualization. Success metrics include reconstruction quality, simulation accuracy compared to traditional physics engines, and end-to-end system latency.

This research advances the field of digital twins by demonstrating how modern ML techniques can enable more responsive and accurate virtual representations of physical environments. The resulting framework will provide a foundation for future work in human-AI collaboration and interactive simulation.

Prerequisites:

  • Python
  • Mathematical background preferred

Learning outcomes:

  • Design and implement an end-to-end ML pipeline that combines computer vision, 3D reconstruction (Gaussian Splatting), and physics simulation (Graph Neural Networks) to create interactive digital twins
  • Evaluate system performance through empirical benchmarking, focusing on reconstruction accuracy, simulation fidelity, and real-time latency requirements

Mentor(s):

  • Krishna Kumar, Assistant Professor, J. Neils Thompson Centennial Teaching Fellowship in Civil Engineering
  • Hassan Iqbal, Postdoctoral Fellow
  • Cheng-Hsi Hsiao, Graduate Student

CosmicAI

The mission of the NSF-Simons AI Institute for Cosmic Origins is to develop AI technologies for astronomical research to advance our understanding of the cosmos and accelerate the pace of new scientific discoveries. The mentor(s) will work with the REU and members of the CosmicAI project to find an astronomical dataset that can be used to help in training of the GenAI models being developed.

Prerequisites:

  • Experience with Python
  • Some familiarity with AI and with astronomical data

Learning outcomes:

  • Experience working with a diverse group of experts
  • Integration of concepts from different disciplines to enable new research outcomes

Mentor(s):

  • Niall Gaffney, Director, Data and AI Directorate, TACC
  • Luke Smith, Research Associate, Scalable Computational Intelligence Group, TACC

Digital Porous Media

The climate crisis and its human impact have created an urgent need to curate and analyze a wide range of geosciences data. Characterizing geomaterials in terms of their microstructural and transport properties is crucial for Earth system understanding and for sustainable resource management. The structure and composition of soil play a foundational role in plant, microbial, and agricultural ecosystems. Deeper below the soil, rock microstructure and mineral heterogeneity are critical to understanding fluid/solid reactions to improve groundwater resources management, carbon sequestration, rare-earth mineral recovery, and contaminant transport. Modern 3D imaging provides a window to the microstructure of soil and rocks; recent advances in machine learning and simulation can improve our understanding of how these materials influence the world around us.

This project will create cloud-based open science tools for visualization, analysis and simulation of geomaterial image data through a repository, Digital Porous Media Portal (DPMP): https://www.digitalporousmedia.org/.

Prerequisites:

  • Programming experience in Python would be a plus

Learning outcomes:

  • Work on a task using dataset(s) from Digital Porous Media Portal
  • Create a Python tool for analyzing some relevant part of the data
  • Produce a presentation detailing their findings and their analysis

Mentor(s):

  • Maša Prodanović, Frank W. Jessen Professor in Petroleum Engineering and Associate Department Chair
  • Bernard (Bernie) Chang, Postdoctoral researcher

Dynamical Systems Lab

Our group studies the prediction and control of chaos. What makes some physical systems easy to predict, while others are harder? The motion of the planets was deciphered by physicists centuries ago, yet the motion of a sloshing cup of coffee remains difficult to predict, even on advanced supercomputers. Our REU student will work on simulating and controlling turbulent fluid flows, with a particular emphasis on creating large-scale turbulent simulations. We plan to use large-scale machine learning models to predict and control the evolution of these turbulent flows.

Prerequisites:

  • Python (numpy, jupyter)

Learning outcomes:

  • Learn about turbulent flows and the science of chaos
  • Learn how to create stable large-scale simulations of complex natural phenomena

Mentor(s):

  • William Gilpin, Assistant Professor, Physics
  • Carson McVay, Graduate Student

Performance Monitoring on HPC Systems

High-Performance Computing (HPC) systems are essential tools for addressing some of today’s most complex scientific and engineering challenges. As these systems continue to grow in scale and complexity, the ability to understand, measure, and optimize their performance becomes increasingly important. This project focuses on the development and application of a performance monitoring tool specifically designed for large-scale HPC applications.

The REU participant will help evaluate how efficiently computational resources are utilized within a scientific application. Activities will include deploying the monitoring tool on one of the world’s largest HPC systems, executing a variety of scientific workflows, collecting and analyzing performance data, and presenting the results. This work plays a vital role in identifying performance bottlenecks and improving resource utilization on state-of-the-art computing platforms.

Prerequisites:

  • Basic understanding of High-Performance Computing (HPC) systems
  • Proficiency in C, C++, Python, and shell scripting
  • Fundamental knowledge of scientific computing

Learning outcomes:

  • HPC Fundamentals: Develop a solid understanding of HPC system architectures, parallel computing concepts, and the challenges associated with running applications at scale. Gain hands-on experience working in a production HPC environment.
  • Performance Analysis and Monitoring: Learn methods and tools for monitoring application performance on HPC systems, identifying performance bottlenecks, and interpreting performance metrics.
  • Programming Skills: Strengthen programming proficiency in C and other languages through practical, performance-focused development tasks.
  • Scientific Workflows: Gain exposure to a variety of scientific workflows and understand how performance monitoring techniques apply to real-world HPC applications.
  • Data Analysis and Visualization: Build skills in processing, analyzing, and visualizing performance data to derive meaningful insights.
  • Research Collaboration: Develop experience working within a research team, collaborating effectively with mentors and peers in a professional research environment.

Mentor(s):

  • Chun-Yaung (Albert) Lu, Research Associate, Texas Advanced Computing Center

Scaling Qualitative Insights: LLM-Driven Analysis for Project MALES

Project MALES produces research and policy briefs focused on the experiences of Latino males across the P-20 educational pipeline. The project also connects UT Austin undergraduate and graduate students (mentors) with middle and high school students (mentees) in local school districts. The REU student will contribute to the development and implementation of an automated system for analyzing large-scale qualitative data (such as interview transcripts and mentor logs) using Large Language Models (LLMs). This work will support existing and future education research by enabling researchers to identify themes and trends more efficiently while maintaining high standards of qualitative rigor.

Prerequisites:

  • Proficiency in Python, particularly for data manipulation (Pandas, NumPy)
  • Basic understanding of Natural Language Processing (NLP) or experience with LLM APIs (e.g., OpenAI, Anthropic, or Llama) - training will be provided
  • Interest in the intersection of social science and computational research

Learning outcomes:

  • Develop a computational pipeline to perform automated thematic analysis on qualitative datasets
  • Design and refine prompt engineering strategies to align AI outputs with social science coding frameworks
  • Evaluate the reliability of AI-generated insights through comparative analysis with human-coded data
  • Gain experience utilizing TACC resources for secure text processing

Mentor(s):

  • Rodrigo Aguayo, Associate Director, Project MALES, College of Education
  • Gabriel Jaffe, Research Associate, Scalable Computational Intelligence Group, TACC
  • Stephanie Baker, Senior Evaluation Researcher, Evaluation Services, TACC
  • Yiwen Yang, Evaluation Researcher, Evaluation Services, TACC

REU participants will also be supported by:

TACC Mentors:

  • Anne Bowen, Research Associate, Data Visualization
  • Justin Drake, Research Associate and Assistant Professor, Women’s Health, UT Dell Medical School
  • Kelly Gaither, Deputy Director and Director of Visualization, and Associate Professor, Women’s Health, UT Dell Medical School