Can open-ended evolutionary principles be used to automate design of structure and control? Can bodies and brains evolve together, stimulating and constraining each other, to yield new artifacts and concepts not programmed in by human engineers? Despite its tremendous potential value, open-ended design automation remains a largely undressed challenge in both AI and Engineering. Design represents a large class of problems that require synthesis. But straightforward composition quickly becomes intractable, and so more elaborate approaches must be found. This talk begins describing a set of experiments in which machines evolve and replicate into reality, thereby establishing for the first time a complete evolution step in physical design. The machines are comprised of only bars, linear actuators and control neurons. Both virtual and real machines evolved for locomotion will be shown, as well as for other tasks. The key hypothesis is that in order to permit scaling to higher complexities, the evolutionary process must be able to discover and reuse knowledge. Based on this idea, new generation of machine has been produced. Some new theoretical results on the emergence of modularity will also be presented.

Time permitting, the speaker will also briefly present results from his Ph.D work on the reconstruction of a 3D object from a single freehand sketch, based on a geometric-correlation machine learning approach.

Thursday, May 3, 2001 @ 4:00 in Duncan Hall 1070
A reception will precede the talk at 3:30pm in DH 3076.

About Hod Lipson

Hod Lipson received the BS degree in 1989 and the Ph.D. in 1998 from Technion. He is presently a Postdoctoral Researcher at Brandeis University, and a Lecturer and Visiting Scholar at Massachusetts Institute of Technology.

Lipson's research interest is in the area of computational design - at the intersection of Mechanical Engineering and Computer Science. He is interested in understanding the design process and emulating it computationally. He looks at both the human design process and at natural evolutionary design as two sources of inspiration and builds working systems to test his theories.

Hod Lipson is a faculty candidate.