Steven Roche

MIT–WHOI Joint Program  |  Reinforcement Learning · Underwater Robotics · Robotic Perception

Photo of Steven Roche
MIT AeroAstro · WHOI
rochesh [at] mit [dot] edu

I am a PhD student in the MIT–WHOI Joint Program, based in MIT AeroAstro's Aerospace Controls Laboratory under Prof. Jonathan How and working closely with WARPLab at Woods Hole under Dr. Yogesh Girdhar. I am supported by a DoD SMART Scholarship.

My research aims to make autonomous underwater vehicles capable, robust field instruments for ocean science, with a focus on coral reefs. I work on reinforcement learning for vehicle control, closing the sim-to-real gap with high-fidelity physics such as computational fluid dynamics, and on robotic perception for autonomous sampling, including environmental DNA collection on reefs.

Before MIT, I was a Gabelli Presidential Scholar at Boston College, where I earned two degrees, a B.S. in Computer Science and a B.S. in Mathematics. There I worked on graph algorithms (min-max correlation clustering, my Scholar of the College thesis) and, as a NOAA Hollings Scholar, on machine learning for GPS-denied magnetic navigation. I have also contributed to robot learning for quadruped locomotion and to whole-brain voltage imaging in larval zebrafish.

news

Sep 2026 Preprint of Feedback-Modulated Harmonic Policies for Quadruped Locomotion posted to arXiv. Paper
Aug 2026 Voltage imaging of neurons distributed across entire brains of larval zebrafish published in Nature Methods. Paper
Jul 2026 Preprint of CORL-AUV: CFD Oriented Reinforcement Learning for Autonomous Underwater Vehicles posted to arXiv. Paper / Project
2026 Received a WHOI Ocean Ventures Fund award for Autonomous eDNA Sampling for Coral Reefs.
Feb 2026 New blog post: how to find and fix a leaky end cap on an underwater robot.
2025 Started graduate school in the MIT–WHOI Joint Program (MIT AeroAstro) as a DoD SMART Scholar.

selected publications

  1. Feedback-Modulated Harmonic Policies for Quadruped Locomotion teaser
    Feedback-Modulated Harmonic Policies for Quadruped Locomotion
    Yixuan Jia, Steven Roche, Jonathan P. How
    arXiv preprint arXiv:2609.17946, 2026
    arXiv PDF BibTeX Abstract
  2. Voltage imaging of neurons distributed across entire brains of larval zebrafish teaser
    Voltage imaging of neurons distributed across entire brains of larval zebrafish
    Zeguan Wang, Jie Zhang, Panagiotis Symvoulidis, Wei Guo, Davy Deng, Adam Amsterdam, Lige Zhang, Takato Honda, Steven Roche, Matthew A. Wilson, Edward S. Boyden
    Nature Methods, 2026
    DOI BibTeX Abstract
  3. CORL-AUV: CFD Oriented Reinforcement Learning for Autonomous Underwater Vehicles teaser
    CORL-AUV: CFD Oriented Reinforcement Learning for Autonomous Underwater Vehicles
    Steven Roche, Milo Van Mooy, Nathan McGuire, Levi Cai, Jonathan P. How, Yogesh Girdhar
    arXiv preprint arXiv:2607.09557, 2026
    arXiv PDF BibTeX Abstract
  4. Min-Max Correlation Clustering via Neighborhood Similarity teaser
    Min-Max Correlation Clustering via Neighborhood Similarity
    Nairen Cao, Steven Roche, Hsin-Hao Su
    European Symposium on Algorithms (ESA), 2025
    arXiv PDF DOI BibTeX Abstract
  5. AI-Driven Magnetic Navigation: Predicting Position from Anomalies teaser
    AI-Driven Magnetic Navigation: Predicting Position from Anomalies
    Steven Roche, Richard Saltus, Manoj Nair
    AGU Annual Meeting (AGU24), 2024
    PDF Poster BibTeX Abstract

All publications →

outside the lab

When I'm not in the lab, I run trail ultramarathons, including races put on by the Trail Animals Running Club (TARC):

Steven at the TARC Fall Classic 50 mile Steven with family at the TARC Fall Classic aid station Steven with a friend at the TARC Fall Classic