Keith's Research

Overview

My work is at the intersection of intelligent systems, robotics, and software systems research. I want to understand the fundamental trade-offs concerning computation, communication, sensing, mobility, and manipulation in robot systems architectures. I have collaborated with systems and robotics researchers on problems involving educational robotics, robot teams, sensor networks, energy-aware robot software systems, and distributed real-time systems. I am an experimentalist, primarily focusing on system building and empirical evaluation as a means for validating theoretical hypotheses. Much of my work has been about managing cost and complexity in robot systems. Managing cost and complexity is of paramount importance for both multi-robot systems research and educational robotics.

Just as special purpose mainframes grew into general purpose personal computers, special purpose industrial robots are evolving into more general purpose personal robots. Although planetary rovers and car assembling robot arms come immediately to mind, traditional software systems are becoming increasingly embodied and situated in their environment -- they are becoming robots. Intelligent systems and autonomous robotics techniques will become necessary ingredients for these systems. We will need to integrate abstractions for uncertainty, mobility, and adaptation into system designs and development tools. I want to explore how intelligent systems and robotics advances can be integrated into our everyday computing infrastructure. Likewise, I want to apply the lessons learned from computing systems research, in terms of software systems and architecture, to robot computing systems.

IPRE - Institute for Personal Robots in Education

This project aims to apply and evaluate robotics as a context for computer science education and is a collaboration between Georgia Tech, Bryn Mawr College, and Microsoft Research. I have helped design and implement various robot platforms for teaching introductory computer science. I taught the class in Fall 2007.

Gnats

We considered how a heterogeneous system composed of many small, embedded, communication nodes and mobile robots equipped with sensors and manipulators could accomplish tasks such as navigation, coverage, and multi-robot foraging. We developed techniques for physical path planning using a real network of embedded nodes. The idea was to use a pervasive network of minimal computing, communicating, and possibly sensing devices to plan paths for mobile robots. Neither the network nodes, or the mobile robots, need to know their positions or build any kind of map. In this type of physical path planning, the embedded nodes act as vertices and communication links represent unit-cost edges. For this study we implemented a new hardware platform, the Gnats. The simplicity of the platform made it inexpensive (less than $30), allowing us to build and experiment with a large number of devices. We get some of the benefits of a robot swarm (pervasiveness) but in a much cheaper, more manageable way than with a fully mobile swarm.

MORPH

We investigated techniques for energy-aware distributed robot software systems. As autonomy increases in robot systems, energy usage for computation will constitute a substantial portion of total energy consumption. In addition, as we look to teams of robots, system energy consumption is increasingly impacted by communications across team members. We developed a system for analyzing system-level energy behavior of autonomous multi-robot systems. We considered the energy implications of a heterogeneous team of mobile and immobile robots conducting a search and rescue mission. This work created a system-level model of the energy behavior of robot software that might be used in a search and rescue mission. First, we created models of the computation and communication energy behavior of the software and hardware. Then by employing distributed computing techniques, the lifetime of mission could be prolonged.

Robocup

As a member of our legged robot soccer team, I competed in 2003-2005 U.S. Opens. I ported the Lua and Ruby programming languages to the Sony AIBO platform, resulting in a more dynamic development environment. In addition, I worked on individual and team robot behaviors. I also advised teams of undergraduates in completing lab projects using the Sony AIBO robots in an Intelligent Perception and Robotics course in the spring semester of 2004.

MARS-2020

We conducted research on multi-robot teams of unmanned air and ground vehicles as part of the DARPA sponsored Mobile Autonomous Robot Software (MARS-2020) initiative. I investigated the use of wireless communication models in multi-robot behaviors and planning. I implemented various communication models inside the MissionLab multi-agent mission specification and simulation environment.