CS 665 : Practicum in Cognitive Architecture for Robots

Instructor : Paul Cohen

Meeting time: 1:30 - 4pm Fridays, Gould Simpson 813

This is a graduate course. Advanced undergraduates may participate with permission of the instructor.

Purpose: In the past it was technically hard to get robots to do anything, and the idea of building a "complete AI" on a robot platform was daunting. It is still challenging to work with robots, but the available software, simulators, and high-quality sensors give me hope that students could not only learn the principles of cognitive architectures but also do useful work to extend the perceptual and cognitive capabilities of a robot. This course will be a practicum: Students will participate in teams to develop aspects of a complete intelligence, such as planning, vision, social interaction, language, and the conceptual system that underlies all else.

Format: One lecture/discussion each week. Self-organized team meetings throughout the semester. Design reviews for projects at three points in the semester. Final demonstrations of projects during finals week.

Grading: The course is a practicum, so the principal grading criterion is participation in projects to develop capabilities for the robot.

Getting Started: A startup guide to set up ROS in your education lab accounts is here. A Google Code page wil be used for most of the technical parts of the course.

Acknowledgments: The hardware and software stack for this course was developed with astonishing speed by Ian Fasel, Daniel Hewlett, Antons Rebguns, Cody Jorgensen, Anh Tran, Jeremy Wright and Cara Slutter. Some of these people were supported by the Defense Advanced Research Projects Agency (DARPA). We are grateful to Willow Garage for ROS (the Robot Operating System) and their attentive and expert support as we developed the software stack.

Resources (in progress)

Lectures and Lecture Notes

Complete Intelligence: Why We Are Doing This (P. Cohen)
Tasks and Abilities. (P. Cohen)
A Gentle Introduction to ROS - Part 1 (D. Hewlett)
Cognitive Architectures - Part 1 (P. Cohen)
Tasks, Abilities and Term Projects. (P. Cohen)

Video clips and online demos

Readings

General:

P. Langley, J. Laird, S. Rogers. Cognitive Architectures: Research Issues and Challenges.
P. Cohen. Against Sophistication.
P. Langley, D. Choi. A Unified Cognitive Architecture for Physical Agents.
W. Kerr, P. Cohen, Y. Chang. Learning and Playing in Wubble World
Oliver G. Selfridge. Pandemonium: A Paradigm for Learning. Thanks to Matt Brockman for tracking this down.
Daniel Corkill. Blackboard Systems.

Concepts:

Robert St. Amant, Clayton T. Morrison, Yu-Han Chang, Paul R. Cohen, and Carole Beal. An Image Schema Language. The International Conference on Cognitive Modeling. 2006. Trieste.
P. Cohen, C. Morrison, E. Cannon. Maps for Verbs: The Relation Between Interaction Dynamics and Verb Use.

Learning:

Y. Chang, C. Morrison, W. Kerr, A. Galstyan, P. Cohen, C. Beal, R. St. Amant, T. Oates. The Jean System.

Other resources

The Soar Homepage. Soar is one of the most extensively developed and applied cognitive architectures.
The ACT-R Homepage. ACT-R may be thought of as a framework for modeling human cognition.
A somewhat dated web site on comparisons between cognitive architectures from John Laird's group at the University of Michigan.
More up-to-date site maintained by A. Samsonovich
Synopsis of Minsky's Society of Mind idea, by Push Singh. Video lecture by Minsky on The Emotion Machine, a followon to the Society of Mind.
"Feature by feature comparison of cognitive architectures" compiled by Alexei Samsonovich and Andrea Stocco

Terms and Concepts you should understand before we are done...

Ability, Affordance, Agent, Asynchronous, Cognitive Architecture, Complete Intelligence, Concept/Category, Control, Deliberative/Reactive, Development, Focus of Attention, Goal, Grounding, Image Schema, Knowledge, Knowledge Level, Learning, Meaning, Memory, Perception, Plan, Reasoning, Representation, Shared Attention, Situated,Social Cognition, State, State Space,

Helpful Instructions

To get started with ROS, run some demos, and generally get familiar with the code environment, check out Daniel Hewlett's instructions.

Team Pages

Presentations on Tasks for the Robot