toru at berkeley dot edu
I am a research scientist at Amazon Frontier AI & Robotics. I build real-world robot learning systems that acquire dexterous, generalizable, and robust skills through real-world and simulated experience.
I received my PhD from Berkeley AI Research (BAIR), advised by Jitendra Malik, and was previously a visiting researcher at FAIR. I completed my BSc and MEng at MIT EECS.
Research
My research focuses on building practical robot learning systems that scale. I am particularly interested in how to combine diverse learning sources to acquire robot skills that are dexterous, generalizable, and robust.
My recent work spans visuotactile bimanual manipulation, sim-to-real RL for dexterous hands, humanoid whole-body control, and preference-aligned manipulation.
I have worked with a range of robotic hardwares and sensors, including bimanual arms, multifingered hands, humanoids, RGBD cameras, force-torque sensors, and tactile sensors.
Publications
Twisting Lids Off with Two Hands
MIMEx: Intrinsic Rewards from Masked Input Modeling
Geometry-Aware Modeling of Rigid Body Physics
Visual Grounding of Learned Physical Models
Teaching
I served as a teaching / lab assistant for the following courses.
- UC Berkeley CS 294-277 Robots That Learn (Fall 2024)
- MIT 6.819/6.869 Advances in Computer Vision (Spring 2021)
- MIT 6.036 Introduction to Machine Learning (Fall 2019)
- MIT 6.863/9.611 Natural Language Processing (Spring 2018)
Miscellaneous
I began my undergraduate studies in physics at The University of Tokyo before transferring to MIT.
Outside of research, I enjoy reading philosophy, learning physics, and making art.
My old personal website is here.