Research

 Webspace of Joel Pfeiffer

Home 

Personal

Research

Publications

Teaching

Classes

Links

Curriculum Vitae

Contact

Interests

My research interests lie at the intersection of Robotics and Machine learning, and lately have been leaning towards path planning.  Most recently, I have worked with planning in the Image Space as opposed to Cartesian planning, although I have implemented more standard approaches in Cartesian planning.

I've been branching out some as well, working on segmentation through various means, as well as looking at Reinforcement Learning with Contact Relative Motions, applied to grasping.  One of the more random interests of mine is music classification, mostly because I'm interested in how to deal with such large amounts of data.


Projects

DARPA Learning Applied to Ground Robots (LAGR) -  I tested Cylindrical planning on the DARPA LAGR robot, working with improving navigation methods in dynamic, outdoor domains.  Most of my recent work has been using fast planning techniques such as Rapidly-exploring Random Trees to try and quickly plan to multiple goals specified by a global planner and choose the best of the returned paths to navigate towards.  I've also had to learn how to deal with elements; surprisingly it gets cold in Colorado.

One of my favorite projects to work on was the Science Crew Operations and Utility Testbed (SCOUT).  As you can see to the right, they actually let me drive it around, which I was pretty stoked about.  I didn't even crash!  My projects with SCOUT involved building a weather station, working with some GPS code, and coming up with a power monitoring system.  Sadly, SCOUT has been replaced by Chariot, which is pretty awesome in its own right, so hopefully I'll get to work on it some future tour.

Image Segmentation - Recently, I have been applying clustering techniques to the problem of image segmentation, trying to take very simple (and somewhat inaccurate) segmentation algorithms and combine their results to create a segmentation that is far better.  We used Non-negative Matrix Factorizations for this task. [PDF]

Music Genre Classifier - For my high dimensional data class, we looked at various ways of classifying music by genre.  I found that the distance matrix was the hard part to come up with.  I've been working on this project some since we made this presentation; perhaps someday I'll get around to putting it up.  Presentation [PDF]

For graphics I made an aquarium!  It started as a sim for a humanoid robot, but I got carried away trying to make it look cool and never really got to that part.  It has some simple code for waves and bubbles, and decent texturing/transparency examples.  Project [7 MB]

Path Planning Comparisons - While at Johnson Space Center, I was tasked with comparing the A* and D* path planning algorithms in various conditions (static, dynamic, indoor, outdoor) with an iCreate Roomba.

Procedure Reference Language - The Procedure Reference Language (PRL) is a XML based language developed at Johnson Space Center to run sequences of commands that is easily programmable through a designer.  The PRL Executor takes the XML generated by the designer and interfaces with the device, executing the commands found in the XML, checks telemetry data from the device, and evaluating conditions on whether or not to continue.