There are two fundamentally different models for building intelligent machines. Using the first strategy (Top-down), you decide what capabilities you want your robot to have and you write a sophisticated computer program (or programs) that 'builds in' all of those cognitive (mental) abilities from the outset. The result is typically a 'logical reasoning' engine that imposes control over the entire robot. The second strategy (Bottom-up), requires no such 'central control program'. Instead many simple (even 'stupid') mechanisms/programs are integrated into one machine. As these many simple devices begin to work in concert, they can produce behavior that is remarkably complex, even 'intelligent'. Our main efforts have thus far been in top-down robot design, but we are beginning a new project in bottom-up robot development. Which one (if either) is the best model for human and animal cognition is a matter of great controversy.
Building Robots From the Top . . Down
Students researchers working on the Iris Project continue to give Iris more and more capabilities. Iris has robotic arms for picking up objects and for drawing letters, a 'learning module' for learning how to win at Tic Tac Toe, a video camera and neural net software for vision, a text-to-speech program for vocalizing and more.
Bottom-up Robotics Projects:
Creating Autonomous Agents
We are ready to launch a new combined artificial life and robotics project which takes a Bottom-up approach. Pete Mandik has offered a proposal for Phase I of the Bottom-up project. We are looking for other teachers and students who would like to participate in this project with us.