Matthew Bronars

I am a first year PhD student in the Robotics Institute at Carnegie Mellon University. I am advised by Katerina Fragkiadaki and my research focuses on robot learning and reasoning, especially for large vision-language action models.

Previously, I recieved my Masters in Computer Science from Georgia Tech (2024) where I was advised by Danfei Xu. I received my BS in EECS and Mechanical Engineering from UC Berkeley (2022). I have also spent time as a Computer Vision intern at the Schlumberger-Doll Research Center (2021) and as a Machine Learning Scientist at Symbotic (2023)

Email  /  Resume

My research goal is to improve the quality of life for all people by enabling embodied AI agents to safely and autonomously opperate in our human-centric world. To this end, I work on deep learning for robotics with a focuses on offline policy generation and data driven approaches to human robot interaction. I am interested in learning from large-scale, unstructed, offline datasets as these are the most abundant and accessible sources of data. Developing algorithims that learn effectively from these datasets is a key step towards deployable robotic agents. I am also interested in how we can leverage humans to help robots gather data, learn new skills, and contend with uncertainty. Algorithms that allow for safe and effective human robot collaboration even in the face of out of distribution data are key to the success of future robotic agents.

Teaching Assistant:

• [Fa 2023] Deep Learning - Georgia Tech CS 7643 (assignment I created on generative models)
• [Sp 2024] Deep Learning for Robotics - Georgia Tech CS 8803 DLM

• Machine Learning - UC Berkeley CS 189 (A)
• [Fa 2022] Artificial Intelligence - Georgia Tech CS 6601 (A)
• [Fa 2022] Computational Data Analysis - Georgia Tech CSE 6740 (A)
• [Sp 2023] Deep Learning - Georgia Tech CS 7643 (A)
• [Sp 2023] Human Robot Interaction - Georgia Tech CS 7633 (A)
• [Fa 2023] Machine Learning with Limited Supervision - Georgia Tech CS 7647 (Current)

Keywords: Imitation Learning, Large Scale Robotic Datasets, OOD Generalization

TL;DR: We develope a simulation framework for generating and evaluating common sources of variation in robotics. Through our analysis, we uncover the types of diversity that should be emphasized during future data collection and best practices for retrieving relevant demonstrations from existing datasets.

Keywords: Diffusion Modeling, Safety Analysis, Imitation Learning

TL;DR: We develop a provably safe method of diffusion policy execution by integrating future state prediction and a reachability-based safety filter.

Keywords: Human Robot Interaction, Diffusion Models,Imitation Learning

TL;DR: By leveraging diffusion model guidance, Legibility Diffuser is able to clone the most legible trajectories from a dataset of multi-modal, multi-task human demonstrations.

Video

Keywords: Imitation Learning, Preference Learning, Manipulation

TL;DR: Learning to Discern (L2D) is an imitation learning framework for learning from suboptimal demonstrations. By training a quality evaluator in a learned latent space, L2D can generalize to new demonstrators given only a small subset of labeled data.

Keywords: Generative Modeling, Human Robot Interaction, Learning from Demonstrations

TL;DR: We introduce Generative Legible Motion Models (GLMM), a framework that utilizes conditional generative models to learn legible trajectories from human demonstrations.

Poster

Keywords: Computer Vision, Image Segmentation, Tracking

TL;DR: Trained a U-Net CNN to segment stem cells and wrote a python script to track them through the timelapse.

Keywords: Computer Vision, Transfer Learning

TL;DR: Designed transfer learning pipeline for implementing cable tracking and damage detection on multiple well sites.

*US Patent Pending

Keywords: Mechatronics

TL;DR: Designed, built, and programmed a pasta making machine for automated food production.