Current Work:

Past Projects:

Snakes:

Active contour models, or snakes, have many applications, most notably in visual tracking. Some videos of my edge-seeking snakes working on simple shapes can be found here. While most work with snakes deals with processing visual data, I designed a mobile robot tracking system using snakes to process only laser data. One avenue that I might like to pursue would be to use snakes on a combination of laser and vision data to increase robustness.

Assisted Helicopter Operation:

On the way to fully autonomous aerial vehicles, this work aimed at assisting helicopter pilots (or teleoperators, in the case of remote-controlled vehicles) by providing helpful navigation information on the operator's HUD and, for some tasks, taking over one or more input control degrees of freedom.

This kind of work must be tested in simulation before moving to the real world, because even small mistakes could cause catastrophic failure in an aerial vehicle. The leading aerial robot simulation software is useful in only large, outdoor environments, but these ideas can be easily applied to miniature helicopters in indoor environments.

For this reason, I designed and implemented a program to simulate and visualize experiments involving aerial vehicles. Like the Gazebo simulator, my system uses ODE as its basic physics engine and OpenGL for visualizations. The big difference is that my system allows the user to input aerodynamics models for specific aerial vehicles for precise, individual simulation results. The simulator allows views from any point in space, including any point attached to vehicles, which translates directly to stationary and mounted cameras so that the simulation results may be directly tested in on real-world platforms.

PPRK Wall Mapping:

The Palm Pilot Robot Kit (PPRK) is the little robot in the picture on my home page. It has three servos that are modified to act as simple motors and three IR sensors, one pointing outward on each side of its body. It is a relatively inexpensive platform, and consequently, its sensing and actuating abilities are rather limited. However, if some complex, intelligent, useful behavior could be achieved with this robot, we could save a lot of money and other resources.

The goal of this project was to use an inexpensive, sensor-limited platform to create accurate maps. I developed a novel pivot method that allowed the robot, which has no built-in odometric capabilities, to map a large section of a hallway in the computer science building with an accuracy of about 95%!

Competitive Co-evolution for Learning Teamwork:

This project was a twist on the classic predator prey evolutionary scenario. By evolving teams of robots rather than individuals, we hoped to show that it is possible to develop cooperative behavior through co-evolutionary processes.

Although we ran into severe technical difficulties that limited the extensive testing that we initially hoped to conduct, preliminary results were encouraging. Due to time constraints and lack of computing resources, we had to use relatively small population sizes and were unable to run simulations for a significantly large number of generations. Nevertheless, single predator robots began to exhibit fairly consistent intelligent behaviors around generation 20, and what could have been the beginnings of cooperative behaviors began appearing, as well.

Life Outside the U