I'm a 2nd year PhD student in Computer Science at Purdue University's Computational Motion, Manipulation, and Autonomy Lab, advised by Zachary Kingston.

In my research I study one of the most basic questions in robotics: how can we make robots that are aware of uncertainty and act safely in the world despite it? Humans do this naturally, but many of today's robots, even those driven by large-scale AI systems, make confidently wrong predictions in new situations, sometimes with costly consequences. My approach combines generative modeling, probabilistic programming, and Bayesian inference to help robots learn what they don't know.

Previously, I was a research engineer at IBM, working with the MIT-IBM Watson AI Lab on DARPA's Machine Common Sense program. I also built ModPPL, a dynamic probabilistic programming language in Rust.

At MIT, I completed my undergraduate degree in Computer Science and Engineering and in Physics, and my M.Eng. in EECS. There, I was advised by Vikash Mansinghka in the Probabilistic Computing Project, where I worked on probabilistic programming for 3D scene perception. As an undergraduate, I researched models of intuitive physics in the Computational Cognitive Science Group.

News

Publications

  • Sampling-based Model Predictive Control as Sequential Monte Carlo Inference
    Austin Garrett, Muyang Yan, Zachary Kingston
    Under review, IEEE Robotics and Automation Letters (RA-L), 2026
    code
  • Barrier Functions Enable Safety-Conscious Force-Feedback Control
    Charles Dawson, Austin Garrett, Falk Pollok, Yang Zhang, Chuchu Fan
    arXiv preprint, 2022
    arXiv
  • 3DP3: 3D Scene Perception via Probabilistic Programming
    Nishad Gothoskar, Marco Cusumano-Towner, Ben Zinberg, Matin Ghavamizadeh, Falk Pollok, Austin Garrett, Joshua B. Tenenbaum, Dan Gutfreund, Vikash K. Mansinghka
    NeurIPS, 2021
    arXiv
  • Infrastructure for Modeling and Inference Engineering with 3D Generative Scene Graphs
    Austin Garrett
    M.Eng. Thesis, MIT, 2021
    code

Software

  • PFMP: Particle Filtering Motion Planning

    Sampling-based MPC as sequential Monte Carlo inference, unifying MPPI, CEM, and related methods as a single SMC kernel. In JAX and MuJoCo MJX.

    code
  • ModPPL

    Experimental modular probabilistic programming language in Rust, with modeling and inference separated by a generative function trait interface.

    codevideo
  • PointCNN

    PyTorch implementation of PointCNN for point-cloud learning.

    code

All repositories →

Writing