PhD Oral Proposal: Daniel Lessin, GDC 5.516

Contact Name: 
Lydia Griffith
Date: 
Jun 14, 2013 10:00am - 10:00pm

PhD Oral Proposal:  Daniel Lessin

Date: Friday, June 14th
Time: 10:00 am
Place:  GDC 5.516
Research Supervisor:  Don Fussell

Title:  Open-Ended Behavioral Complexity for Evolved Virtual Creatures

Abstract
The development of artificial life has been described as
"extending the horizons of empirical research in biology
beyond the territory circumscribed by life-as-we-know-it and
into the domain of life-as-it-could-be".
This proposed dissertation uses the tools of
artificial life to apply the same type of extension to the
field of biologically derived content. Moving from
evolved-creature-content-as-we-know-it (in the natural world)
to evolved-creature-content-as-it-could-be (in the virtual
world) offers a myriad of potential advantages, including
novel morphologies, on-demand production, new opportunities
for human input and control, and the ability to be deployed in
virtual environments.

In 1994, this endeavor got a powerful kick-start from Karl
Sims' landmark publication on evolved virtual creatures
(EVCs), which introduced a
form of artificial life in which the co-evolution of
morphology and control for physically simulated creatures is
used to produce novel and compelling virtual creature content.
In the years that followed, however, despite a number of
interesting extensions to Sims' work, a large category of EVC
content has remained completely unexplored.  Despite the
promise of great potential value, there has been no clear
increase in the behavioral complexity of evolved virtual
creatures beyond the results demonstrated in Sims' original
work.

This proposal presents
an open-ended method to break this barrier, making use of
high-level human input in the form of a syllabus of
intermediate learning tasks--along with mechanisms for
preservation, reuse, and combination of previously learned
tasks.

For completed work, this method (named ESP for its
three components: encapsulation, syllabus, and
pandemonium) is implemented and employed to evolve a
virtual creature with behavioral complexity that approximately
doubles the state of the art, thereby taking the first new
steps in almost two decades into what may well be vast
and rich new territories of
evolved-creature-content-as-it-could-be.

For proposed work, improvements are described to increase the
system's yield and flexibility, which are currently limited by
self-imposed constraints originally applied to make positive
initial results more likely.  If successful, these
improvements would mark a significant additional step toward
the practical application of this new technique.