CS 395T: Robot Learning
- Fall 2001. MW, 10:30 am - 12:00 pm.
- TAY 3.144
- Unique #51920.
- Professor Benjamin Kuipers
- Office hours: MW 1:30 - 2:30 pm. TAY 4.130C.
In this course, we will study the problem of how an agent can learn to
perceive its world well enough to act in it and make
reliable plans. Studying this problem in the context of physical
robots forces us to confront the continuous nature of sensory input,
of action, and of the environment in which the agent exists. Our goal
is to determine how the agent can discover useful abstractions from
continuous sensorimotor experience, to build symbolic knowledge
representations describing its world.
This problem is of practical importance, because the robots of the
future will have increasingly rich sensory systems. It will be
necessary for them to learn from their own experience how their
sensors respond to the world and how the world responds to the actions
they take. The meanings of symbols in their knowledge representation
are then "grounded" in the nature of their sensorimotor interaction
with the environment. In addition to its practical importance, this
problem is central to the philosophical problem of whether artificial
intelligence is possible at all (Searle, Harnad).
We will start from a theory of spatial exploration, mapping, and
navigation in unknown environments to define a target for the agent's
learning process. We will examine previous work (by Gary Drescher,
David Pierce, Wei-Min Shen, Paul Cohen, Luc Steels, Rob Schapire, Pat
Langley, etc.) on learning from an unknown sensory world.
Experiments and assignments will be done on simulated and physical
robots. One of our physical robots is a prototype of an intelligent
This is a research seminar, intended first to bring you to the state
of the art, and then to help you do a project and paper of publishable
quality. There will be a significant amount of reading of research
papers that will be handed out.
There will be several mathematical and programming assignments, a term
project and a presentation.
Each class member will select a topic and present the material to the
class. Each topic will have an associated reading that the entire
class will read, but the presenter is responsible for finding and
reading additional material, becoming an expert in the area, creating
an illuminating example to present, and leading a discussion.
Plan to give a 25 minute presentation, followed by questions and
discussion of the value and importance of the material presented.
- What is the problem? Why is it important? Why should the reader care?
- What assumptions are being made?
- How does this method work? (Provide an intuition that will guide
the hearer as they read the technical details.)
- What are the strengths and limitations of this approach?
- How can you evaluate the benefits?
- What are open problems in this area?
- How does this help us?
Where is the gold?
(We may spend multiple sessions on some of these.)
Obviously, it will be helpful to pick a presentation topic that works
well with your term project topic.
- Gary Drescher's schema mechanism.
- Marvin Minsky, The Society of Mind.
- Wei-Min Shen, Autonomous Learning from the Environment.
- Leslie Valiant, Circuits of the Mind.
- Neural-net function learning.
- growing and hierarchical networks (Mark Ring; Fritzke; ...)
- Seong and Widrow on neural nets for optimal control.
- Kohonen, Self-organizing maps.
- Heckerman, Learning Bayesian nets.
- Mixture of experts: Jordan & Jacobs; Hinton; Ghosh.
- Atkeson, Moore & Schaal on locally weighted learning.
- Langley, et al, on the BACON family of scientific discovery methods.
- Unsupervised clustering methods.
- The EM algorithm.
- Leslie Kaelbling, Reinforcement learning and POMDPs.
- Andrew McCallum, The Utree algorithm.
- Two approaches to mapping and localization: Durrant-Whyte vs. Thrun.
- Luc Steels, language learning.
- J. J. Gibson, learning affordances (potential for action).
Each class member will do a term project. You can apply a method we
are learning about to a robot learning problem. Or you can extend an
existing method or develop a new method to solve a problem. Ideally,
your term project will extend the state of the art, and will be
suitable for submission to AAAI or ICRA or some other major conference.
You are encouraged to select a topic that fits well with your other
The textbooks are references that will be valuable in different ways
in this area. The University Coop may not have these in time, so you
might consider buying them from Amazon or Barnes & Noble.
You will also need to do assignments in either MATLAB or LabVIEW.
Make sure you have any documentation you need.
- Gary L. Drescher. 1991. Made-Up Minds: A Constructivist Approach
to Artificial Intelligence.
Cambridge, MA: MIT Press.
(This book is an important landmark on the path we want to explore.)
- Duda, Hart and Stork. 2001. Pattern Classification, Second Edition.
NY: John Wiley and Sons.
(This book is a useful reference for your library.)
- Tom Mitchell. 1997. Machine Learning. Boston: McGraw-Hill.
(This book is a useful reference, and is required for Ray Mooney's
Machine Learning course.)
If you do not already have a background in Artificial Intelligence,
you should invest in the following excellent textbook. It's another
valuable addition to your library, and is undoubtedly available used.
- Stuart Russell and Peter Norvig. 1995. Artificial
Intelligence: A Modern Approach. Prentice-Hall.
Resources (check for updates)
The following resources are from last year's
CS 395T: Intelligent Robotics,
which had a significantly different focus. These links will be updated,
but you might still find them interesting.
Students with Disabilities
The University of Texas at Austin provides upon request appropriate
academic accommodations for qualified students with disabilities. For
more information, contact the Office of the Dean of Students at
471-6259, 471-4641 TTY.