Patrick Grady

Hi! I'm currently a Robotics PhD student at the Georgia Institute for Technology, where I'm advised by Charlie Kemp. I work on computer vision for robots.

Before coming to Georgia Tech, I did my undergrad at Duke in computer science and electrical and computer engineering. I got to lead Duke Electric Vehicles, a team of undergraduates building ultra-efficient vehicles. During my tenure as president, we set the world record for Most Fuel-Efficient Vehicle at 14,573 MPG, and Most Efficient Electric Vehicle at 27,482 MPGe.

CV  /  Google Scholar  /  LinkedIn

profile photo
Research

My ultimate interest is to make robots that are able to interact with our unpredictable world. In the shorter term, my research is developing computer vision algorithms that produce actionable information for robots.

BodyPressure BodyPressure - Inferring Body Pose and Contact Pressure from a Depth Image
Henry M. Clever, Patrick Grady, Greg Turk, Charles C. Kemp
Under review, 2021

We present a method that infers contact pressure between a human body and a mattress from a depth image under occlusion from blankets. Our approach augments a real dataset with synthetic data generated via a soft-body physics simulation. We introduce a deep network trained on this augmented dataset and evaluated with real data.

ConatctOpt ContactOpt: Optimizing Contact to Improve Grasps
Patrick Grady, Chengcheng Tang, Christopher D. Twigg, Minh Vo, Samarth Brahmbhatt, Charles C. Kemp
Conference on Computer Vision and Pattern Recognition, 2021 (oral)

Physical contact plays a critical role in hand and object grasping. By estimating desirable contact then optimizing the hand pose to achieve it, ContactOpt improves the accuracy and realism of estimated hand and object poses.

Dressing Masked Reconstruction based Self-Supervision for Human Activity Recognition
Harish Haresamudram, Apoorva Beedu, Varun Agrawal, Patrick Grady, Irfan Essa, Judy Hoffman, Thomas Ploetz
Ubiquitous Computing/International Semantic Web Conference, 2020

Human Activity Recognition (HAR) datasets are typically small and limited in variability. We demonstrate that self-supervised masked-reconstruction learning improves the performance of activity classifiers on benchmark datasets as compared to unsupervised techniques.

Dressing Learning to Collaborate From Simulation for Robot-Assisted Dressing
Alexander Clegg, Zackory Erickson, Patrick Grady, Greg Turk, Charles C. Kemp, C. Karen Liu
IEEE Robotics and Automation Letters, 2020

We investigate haptic feedback control for robot-assisted dressing. We use deep reinforcement learning to simultaneously train human and robot policies to accomplish the task while modelling human impairments.

Maxwell A Study of Energy Losses in the World's Most Fuel Efficient Vehicle
Patrick Grady, Gerry Chen, Shomik Verma, Aniruddh Marellapudi, Nico Hotz
Vehicle Propulsion and Powertrain Conference, 2019 (oral)

This work studies Maxwell, our hydrogen fuel-cell powered vehicle. We quantify the energy losses in the vehicle's powertrain with extensive experimentation and simulation. We demonstrate excellent correlation with measured on-track vehicle performance.

Teaching
gt Graduate Teaching Assistant, CS 7643, Deep Learning

Graduate Teaching Assistant, CS 6476, Computer Vision

Graduate Teaching Assistant, ECE 3072, Electrical Energy

Website template by Jon Barron.