Andrew Hundt

Twitter: @athundt Email: ATHundt@gmail.com

PhD, Johns Hopkins University

My goal is to create robots and other learning agents with brilliant adaptability, skillsets, and capabilities that are tailored to respect human Equity, Diversity, and Inclusion (EDI).

Democratizing Access: Natural Cooperation between Diverse Human Populations and Learning Agents

I’m Computing Innovation Fellow with the Computing Research Association (CRA) under National Science Foundation (NSF) support, for a Postdoctoral Fellowship at the Georgia Institute of Technology with Assistant Professor Matthew Gombolay and Regents’ Professor Ronald C. Arkin.

Our objective is to develop human and technical principles, processes, metrics, and algorithms that advance Equity, Diversity, and Inclusion in the development of computational methods for Robotics and AI.

Selection of Papers

See my Google Scholar page for the complete and up to date list because this page is not updated frequently.

“Good Robot! Now Watch This!”: Repurposing Reinforcement Learning for Task-to-Task Transfer - Conference on Robot Learning 2021

Guiding Multi-Step Rearrangement Tasks with Natural Language Instructions - Conference on Robot Learning 2021

“Good Robot!”: Efficient Reinforcement Learning for Multi-Step Visual Tasks with Sim to Real Transfer - Robotics and Automation Letters (RA-L) 2020

SharpDARTS: Faster and More Accurate Differentiable Architecture Search - Preprint 2019

Sets state of the art (SOTA) on the CIFAR-10 dataset and matches SOTA on ImageNet at mobile scale in March 2019. Proposes Differentiable Hyperparameter Search and Cosine Power Annealing.

The CoSTAR Block Stacking Dataset: Learning with Workspace Constraints - IROS 2019

A dynamic robotics dataset which serves as a valuable resource to researchers across a broad range of topics. Proposes HyperTree Neural Architecture Search (NAS) for refining neural network models. CoSTAR Block Stacking Dataset Website

CoSTAR: Instructing Collaborative Robots with Behavior Trees and Vision - ICRA 2017

Defines perception and spatial reasoning abstractions to enable end-users to create robust task plans. Winner of the 2016 KUKA Innovation Award in a competition on flexible manufacturing.

User Experience of the CoSTAR System for Instruction of Collaborative Robots - IROS 2017

A user study showing the CoSTAR flexible manufacturing system is effective for specifying task plans.

Robone Project

I’ve worked on a robotic hip replacement surgical assistant robot.

Generic Robotics Library

Past Employment

Robotics Engineer National Robotics Engineering Center, Robotics Institute, Carnegie Mellon University

Farming Automation: I’ve worked on autonomous orchard and field tractors.

Boost.Container static_vector

Part of Boost C++ Libraries, static_vector, aka varray, is a sequence container like std::vector with the benefits of statically allocated storage as found in std::array. Acknowledgments: Boost.Geometry, Boost.Container

Personal Projects from 2015 or earlier

CalcHundter

A cross-platform calculator app to assist with programming arithmetic and bit operations, made with the use of HCI research and design principles. Download: Windows, Mac

Wii Remote Controlled Robot

In 2007 I built a robot that can be controlled by Wii remote.

CMU Buggy

I was an engineer for Fringe Racing, a team which designs and builds human powered, ultra light carbon composite vehicles. Fringe Racing placed first in the Carnegie Mellon Sweepstakes Design Competition in 2006, 2007 and 2008. Elected Chairman of the Sweepstakes Committee for 2009; led the school wide organization with 13 teams and 250 members.

Inertial Electrostatic Confinement fusion reactor

In High School I researched an Inertial Electrostatic Confinement fusion reactor which achieved plasma.