Marcus Dominguez-Kuhne

I am a PhD student at the University of Southern California (USC) advised by Professsor Gaurav Sukhatme in RESL (Robotics Embedded Systems Laboratory). I work at the intersection of machine learning, robotics, and computer vision. My research focuses on using machine learning techniques such as deep reinforcement learning and imitation learning for robotic applications including perception and manipulation. I am funded by a USC four year fellowship.

I completed my Bachelors in Computer Science at the California Institute of Technology (Caltech), where I worked with Professor Yisong Yue on multi-agent Motion Planning using Imitation Learning. Also I have researched in the UC Berkeley AUTOLab under Professor Ken Goldberg using machine learning to intelligently search for desired objects in shelf environments. Additionally, I have researched in the Stanford Vision and Learning Lab (SVL) under Professor Silvio Savarese using deep reinforcement learning for object retrieval in bins.

Email  /  Resume  /  Google Scholar  /  LinkedIn

profile photo
Research

I'm interested in reinforcement learning, imitation learning, and computer vision for general purpose robotic agents focused on manipulation. Much of my research has focused on the mechanical search problem searching for physical items in bins and shelves and how to use robots in dynamic, home settings to better assist individuals.

Publications & Preprints
Mechanical Search on Shelves using Lateral Access X-RAY
Raven Huang*, Marcus Dominguez-Kuhne*, Jeffery Ichnowski, Vishal Satish, Michael Danielczuk, Kate Sanders, Andrew Lee, Anelia Angelova, Vincent Vanhoucke, Ken Goldberg
* Indicates equal contribution
IROS (International Conference on Intelligent Robots and Systems) , 2021
project page / arXiv / Venture Beat Article

LAX-RAY (Lateral Access maXimal Reduction of occupancY support Area) is a system to automate mechanical search for occluded objects on shelves. LAX-RAY couples a perception pipeline predicting target object occupancy support distribution with a mechanical search policy to sequentially select occluding objects to push sideways to quickly reveal the target.

Visuomotor Mechanical Search: Learning to Retrieve Target Objects in Clutter
Andrey Kurenkov*, Joseph Taglic*, Rohun Kulkarni, Marcus Dominguez-Kuhne, Animesh Garg, Roberto Martin-Martin, Silvio Savarese
* Indicates equal contribution
IROS (International Conference on Intelligent Robots and Systems) , 2020
project page / arXiv

In this work we present a Deep Reinforcement Learning procedure that combines i) teacher-aided exploration, ii) a critic with privileged information, and iii) mid-level representations, resulting in sample efficient and effective learning for the problem of uncovering a target object occluded by a heap of unknown objects. Our appraoch, trains faster and converges to more efficient uncovering solutions than baselines and ablations, and our policies lead to an average improvement for target graspability.

Teaching

Teaching Assistant: Caltech CS/EE 155 [2020], Machine Learning/Data Mining

Teaching Assistant: Caltech CS/EE 155 [2019], Machine Learning/Data Mining

Teaching Assistant: Caltech CS/EE 156a [2019], Learning Systems

Service

Reviewer: IROS 2021


source code