|
Leo Huang I'm a robotics researcher at Berkeley AI Research (BAIR) and the EMBER Center, advised by Prof. Shankar Sastry. I work with Prof. Sastry and Prof. Claire Tomlin on the DARPA TIAMAT program, where our Berkeley and Stanford team placed 2nd in Phase 1. I build manipulation tools for our Spot field deployments. I received my M.S. in EECS from UC Berkeley in 2025, following B.S. degrees in EECS and Bioengineering. Before robotics, I worked on DNA origami placement with Prof. Grigory Tikhomirov (my M.S. thesis) and on magnetic particle imaging with Prof. Steven Conolly. I also worked on vision-language navigation for quadrupeds with Prof. Koushil Sreenath. |
|
ResearchI'm working toward humanoid and mobile manipulators that can safely provide physical assistance to people. I want to make robot adaptation rapid and measurable, teach robots resourceful, creative problem solving, and build whole-body dexterous manipulation with safe learning-based control. My current work covers multimodal spatial memory for object search (FARM) and improving a robot's tools in simulation before field deployment (Double-T DINER). |
News
|
Publications* denotes equal contribution. |
 |
FARM: Find Anything using Relational Spatial Memory
Siming He, Leo Huang, Adam Lilja, Fabio Huebel, Jonas Frey, Marco Pavone, S. Shankar Sastry, Jitendra Malik, Claire Tomlin arXiv, 2026 arXiv A real-time, open-vocabulary object memory that answers queries describing an object by its relations to landmarks and nearby objects. Across 44k queries in 67 indoor and outdoor scenes, FARM improves Recall@5 by 164% over prior methods, and it runs closed-loop on a quadruped. |
 |
Double-T DINER: Deploying Improvements in New Environments via Refined Tool Transfer
Henry Tsai*, Leo Huang*, Denae Cantrell*, Narek Harutyunyan, Hunter Tsai, Maggie Dong, David Jo, Claire Tomlin, S. Shankar Sastry CoRL 2026 - Sim2Real2Field Workshop, 2026 project page A refiner agent edits a Spot robot's pick-and-place tools in simulation, and we test whether the edits that help in simulation also help in real-world field trials. |
 |
Precision Placement of DNA Origami onto Patterned Silicon Wafer Surfaces
Leo Huang M.S. Thesis, UC Berkeley EECS, 2025 (Technical Report UCB/EECS-2025-76) tech report Placed hexagonal DNA origami onto patterned silicon wafers and raised single-origami binding yield 4x by tuning buffer conditions and deposition methods. |
 |
Broadband Untuned Active Cancellation and Phase Correction of Direct Feedthrough Interference Enables 100-fold Improved Magnetic Particle Imaging Detection
Q. Huynh, O. Doyle, K. Guo, A. Swamynathan, L. Huang, J. Garlepp, N. Dambal, J. Bryan, S. Ramdani, C. Saayujya, B. Fung, Y. Lu, B. Fellows, O. Sehl, S. Conolly TechRxiv, 2023 preprint An untuned active cancellation circuit removes direct feedthrough interference in magnetic particle imaging and improves detection 100-fold. |
Projects |
 |
Project Pigeon
1st Place, Bee AI Hackathon, 2025 Voice-command control for a DJI Tello drone, combining open-vocabulary (OWL-ViT) and real-time (YOLO) detection with a PID controller for vision-based closed-loop flight. |
Honors
Teaching
|
|
Website template from Jon Barron. |