Note: a pdf version of my CV can be found here.
My thesis proposal, Articulated SLAM can be found here.
I was co-author on each of these works:
Matt Klingensmith, Michael Koval, Sidd Sirinivasa, Michael Kaess and Nancy Pollard
In this work, we estimate the state of a noisy robot manipulator and under-actuated hand using a Manifold Particle Filter. We extend the filter to higher dimensional state spaces by representing contact manifolds implicitly and projecting onto them (rather than explicitly sampling them).
(ICRA + RA-L 2016. Best computer vision paper finalist.)
Matt Klingensmith, Sidd Sirinivasa, Michael Kaess
We present a method for robot arm state estimation and simultaneous 3D reconstruction using a hand-mounted depth sensor. Unlike other techniques, our method explicitly estimates state in the robot configuration space rather than in the workspace.
Chisel: Real-time 3D Reconstruction Onboard a Mobile Device using Spatially Hashed Truncated Signed Distance Fields
(RSS 2015. Best Systems Paper Finalist)
Matt Klingensmith, Ivan Dryanovski, et. al
In this paper we present an algorithm for creating large scale, high quality volumetric maps using sparse, noisy sensor data on the Google Tango device. By using a novel chunked data structure for storing the volumetric data, we use much less memory than typical approaches, allowing us to map much larger areas.
Klingensmith, Hermann et. al
This paper emphasizes the use of negative space in object reconstruction and matching, using an efficient voxel carving technique. We extended Hermann’s technique to shape reconstruction and matching, and provide a principled approach for object priors from depth images.
(Workshop ICRA 2013)
Klingensmith, et. al
This is a systems paper detailing an extremely fast and efficient technique of correcting robot arm localization errors using noisy depth images. We use the technique to control a robot arm to open a door.
Javdani, Klingensmith et. al.
This paper lays theoretical groundwork for blind, touch-based localization. We use the technique to open a door from blind touches. The touching problem is related to submodularity. This allows us to select actions greedily while maintaining optimality bounds. A particle filter approach is used to localize the door, and ultimately, open it.
Zucker, Ratliff, et. al.
This is a journal paper of the CHOMP planner and its extensions. It was a large collaboration paper between several CMU faculty and students who had used/implemented/extended the CHOMP planner. Due to my work on optimizing the CHOMP planner for ARM-S, I was invited to write a section on collision cost and joint limits.
(Tech Report 2012)
Pivtoraiko, Klingensmith et. al.
This paper presents improvements on CHOMP for real-time replanning. I wrote most of the code used in the test implementation, and ran most of the experiments. The key performance improvement was in replacing the signed distance field to an overlapping compound of distance field volumes. Replanning is scheduled accordingly to minimize the amount of time needed for planning/executing trajectories.
Bagnell, Hebert, Pollard et. al.
This paper describes software/algorithms used in the ARM-S Competition. This is a large collaboration paper of everyone involved in NRECs competition team, which I was on for several years. It turns out that doing manipulation in the “real world” is very difficult. Many of the grasping algorithms from the literature were totally useless under uncertainty, requiring us to develop manipulation strategies that took into account touch and uncertainty reduction.
Dyson Robotics Lab, Imperial College London
Visiting Researcher Spring 2016,
I worked with Profs. Andrew Davison and Stefan Leutenegger on SLAM concepts for robot manipulators, with applications toward automatic robot calibration.
Google/Motorola Advanced Technologies and Projects
Temporary Project Staff Summer 2014,
Worked on Google’s project Tango, a mobile device with onboard IMUs, depth cameras, and visual sensors. My role here was to develop real-time 3D mapping capabilities for the sensor, which culminated in the Chisel application. My main advisor was Joel Hesch. I worked closely with Ivan Dryanovski from CCNY and Simon Lynen from ETH Zurich.
National Robotics Engineering Center (NREC)
Student Intern 2012 – 2013,
Software Engineer Summer 2013
Employed as a student intern (then as a full-time employee) to work on the DARPA ARM-S project. Programmed key parts of the software infrastructure, planning algorithms and high-level behaviors. Assisted with on-site testing.
Software Engineering Intern 2011
Worked with Dr. Gil Jones on the ROS Electric Arm Navigation stack. Made improvements to the CHOMP algorithm allowing planning safely out of collision. Wrote key software (Arm Navigation Wizard and the Arm Navigation Warehouse) which would become components of the MoveIt software framework.
Software Engineering Intern 2010
Developed sensor visualization, testing, and user-interface software during development of the DARPA ARM robot. Assisted in training the DARPA ARM teams, and created a block-placing demo using the robot.
Carnegie Mellon Robotics Institute, 2013 – present
Intended Degree: Doctor of Philosophy
Current QPA: 4.0, Advisors: Sidd Sirinivasa and Michael Kaess
Carnegie Mellon Robotics Institute, 2011 – 2012
Degree: Master of Science in Robotics
Final QPA: 3.9, Advisor: Nancy Pollard
Carnegie Mellon School of Computer Science, 2008 – 2011
Degree: Bachelor of Science in Computer Science, Minor in Robotics
Final QPA: 3.21
Grapevine High School, 2004 – 2008
Final GPA: 3.88. Final Rank: 33 / 476