Philosophy

I see artificial intelligence as an interdisciplinary science, both theoretical and applied. As a theoretical science, artificial intelligence benefits from research carried out in cognitive science, cognitive psychology, neuroscience, and other fields. In turn, artificial intelligence can benefit these fields by suggesting approaches and testing theories in ways not otherwise possible. As an applied science, artificial intelligence can provide intelligent tools for use in other areas and learn from the application of established methods to unique domains.

Collaboration is an integral part of doing artificial intelligence research. Due to its interdisciplinary nature, an artificial intelligence researcher must be willing and able to work with researchers from other fields. Further, specialists in subdomains within artificial intelligence can profit from sharing in each other's research. Such collaboration is useful both to create joint projects and to serve as a mechanism for cross-fertilization between areas.

As with all sciences, artificial intelligence must make use of empirical data. Creating applications for use in the real world not only provides funding opportunities, it gives artificial intelligence researchers a rigorous testing environment.

Interests

I am interested in conducting research in many subdomains within artificial intelligence, including robotics, machine learning, and knowledge-based systems, and in integrating research from these areas into working systems. My background within the field is both broad and deep and I have an affinity for the natural sciences that allows for easy collaboration with researchers in other fields.

Robotics

I am interested in all aspects of robotics, particularly autonomous learning and action, distributed robots and multi-agent systems, and biologically-inspired robotics. I have worked with systems ranging from large commercial systems, such as the Puma-560 manipulator and the ATRV-Jr. all-terrain mobile robot, to miniature robots that we designed and built in our laboratories.

I have shown that learning of simple tasks can take place autonomously on board robots with very little computing power in a brief period of time. This research was carried out first in simulation, using an X-Window-based, graphical user interface of my own design for the display of learning system parameters and robot simulations, then demonstrated using real custom-built robots.

I helped to lead a team to design and construct a novel miniature robot known as the scout - smaller than a soda can - which can roll, hop, transmit video or other sensory information, and act as a communication node in a robotic sensory and communication network. Further, the rugged and cylindrical scout can be deployed by tossing it by hand or shooting it from a scout launcher, which we also designed and built. We designed a flexible software architecture for controlling large teams of scouts and rangers, which are larger robots with more computational power that act as cooperative coordinators of the team.

Using the scouts and rangers and inspired by biological systems, I designed a scenario in which the robotic team carried out a reconnaissance and surveillance mission by executing simple behaviors such as finding doorways and dark areas.

Machine Learning

I am interested in the development of novel machine-learning algorithms and the application of machine learning to areas including robotics and the sciences, particularly biology.

My doctoral research focused on a unique connectionist reinforcement-learning system for learning robotic control behaviors. This system uses principles from what are widely known as "Kohonen's Self-Organizing Topological Feature Maps." Using biologically inspired additions and revisions to these maps I have created an original system capable of rapid learning. The creation of this system brought me my first professional publication and a research contribution award from the Department of Computer Science, both in 1993.

More recently, I have developed a number of machine learning algorithms, including NQ-Learning, NTD-Learning, and Memetic Algorithms. These systems take advantage of generalization and imitation to provide for rapid reinforcement and evolutionary learning. These algorithms are applicable to control and bioscience problems.

Knowledge-Based Systems

I am interested in the creation of original and practical knowledge-based systems and have had the opportunity to do this in the Department of Soil, Water, and Climate at the University of Minnesota. Besides developing conventional expert systems and geographic information systems, I combined these systems, resulting in powerful and useful knowledge engines. The systems were developed to help preserve the environment while improving efficiency for the user.

Experience

As a graduate student, I led a team of students, including graduate and undergraduate students and a gifted high-school student to construct a set of mini-robots for research and demonstration purposes. I sought out and received funding to bring these robots and the most active team members to the American Association of Artificial Intelligence conference in 1994. This was the first participation by the University of Minnesota in the AAAI Robot Competition and Exhibition but the enthusiasm generated by this experience ensured that participation in this event continued in subsequent years.

My experiences in the Department of Soil, Water, and Climate provided me with practical experience in system development and project management. I was involved in all aspects of these projects, from initial meetings with funding agencies and potential users, through developing prototypes and working systems, coordinating beta-testing, and acting as technical support. I am pleased to say that systems I developed are currently in use throughout the state of Minnesota.

As Associate Director of the Center for Distributed Robotics, I combined my leadership skills with my knowledge of system development and project management by helping to lead a large team from academia and industry to develop a heterogeneous multi-robot system. Our team has included eight faculty members and many graduate and undergraduate students and post-doctoral researchers from the Department of Computer Science and Engineering, the Department of Electrical and Computer Engineering, and the Department of Mechanical Engineering, as well as our industry partners at MTS Systems Corporation, Honeywell Laboratories, and Architecture Technology Corporation. It is this team that developed and built the scout and ranger robot system described above. This project has been highly successful, resulting in numerous publications as well as awards, patents, and a great deal of positive publicity for the departments and companies involved, the University of Minnesota, and robotics research in general.

Goals

Short-Term Goals: Building on Existing Research

While the basic research on the scouts is completed and we are moving them rapidly to commercialization, we have plans for additional research to provide the scouts with additional capabilities, such as chaining and wall-climbing, and to provide additional control methods for them.

My new learning methods (NQ-Learning, NTD-Learning, and Memetic Algorithms) show great promise and I plan to test them thoroughly, compare them with existing methods, and demonstrate their usefulness in real-world applications.

I plan to revisit my research on the combination of expert systems and geographic information systems, exploring the theoretical aspects in more detail than was possible in my position with the Department of Soil, Water, and Climate. I also see a potential for developing similar practical systems for many other areas of land-use management.

The use of inexpensive mini-robots and the chance to take part in international events have proven to be great motivators for student participation and I plan to continue to use these.

Medium-Term Goals: Expanding and Integrating Research

I am interested in integrating disparate research on robotics into working systems that will take robots beyond factories and into useful tasks in semi-structured environments. Towards this end I have proposed Bibliobot as a first stage in developing robots for use in commercial environments.

Such robots would need to be largely autonomous and capable of dealing with a wide range of unplanned situations. Coordination of simple learned behaviors can provide a large behavioral repertoire and I am investigating methods that can integrate learning into the behavioral mission decomposition system we developed for our scout and ranger robots.

To make such robots feasible, better sensing is required as well. I have proposed a novel, biologically-inspired model for the integration of multiple low resolution sensors such as sonars and infrared light sensors to create a single sensory field around the robot. Further biological inspiration leads to vision systems with much lower computational costs than those of today.

My research in Memetic Algorithms will lead me to other memetic learning methods including Memetic Programming and Memetic Systems. Further, I am interested in taking learning to more abstract levels, away from low-level control and optimization problems. This will lead to yet another meme-inspired learning system, Imitation-Based Learning, and an integration of memetic learning theory into machine learning theory.

In knowledge-based systems, applications of expert system technology to other information sources (such as digital libraries) could prove very powerful.

Long-Term Goals: Intelligent Agents for the Real World

Systems that can help people in the sciences, engineering, business, politics, and other fields to understand their world better -- this is the first long-term goal of my research. The learning systems, knowledge-based systems, geographic information systems, and integrated systems on which I am working will find application in practical tools for understanding.

Agents that can navigate in the globe-spanning electronic world of the Internet and the Web, not only gleaning information of importance to their users as they explore sites filled with content of many different types, but also learning about this virtual world as they travel, learning about the goals of their users as they interact with them, and presenting the discovered information in meaningful ways -- this is the second long-term goal of my research. It will take an integration of learning with knowledge-use, sensing and reacting with planning and goals.

Robots that can make their way through the day to day world around us; traveling from place to place, even from country to country; dealing with people, animals, machines, and the weather; learning about the world; reacting in real time; helping people to accomplish their goals -- this is the third and most ambitious of my long-term goals. It will take an integration of all of my previous research and more.

These long-term goals are not ones that can be achieved by a single person or a single research initiative. They are shared goals of many of us in the field of Artificial Intelligence and my role will be to do what I can to work with others to achieve them.

Last update: 21 August 2001

© 2000-2001. All rights reserved.