Evolved Virtual Creatures as Content: Increasing Behavioral and Morphological Complexity (2014)
Throughout history, creature-based content has been a highly valued source of entertainment. With the introduction of evolved virtual creatures (or EVCs) by Karl Sims in 1994, a new source of creature content became available. Despite their immediate appeal, however, EVCs still lag far behind their natural counterparts: Neither their morphology nor their behavior is sufficiently complex. This dissertation presents three contributions to address this problem. First, the ESP system, which combines a human-designed syllabus with encapsulation and conflict-resolution mechanisms, is used to approximately double the state of the art in behavioral complexity for evolved virtual creatures. Second, an extension to ESP is presented that allows full morphological adaptation to continue beyond the initial skill. It produces both a greater variety of solutions and solutions with higher fitness. Third, a muscle-drive system is demonstrated to embody a significant degree of physical intelligence. It increases morphological complexity and reduces demands on the brain, thus freeing resources for more complex behaviors. Together, these contributions bring evolved virtual creatures, in both action and form, a significant step closer to matching the entertainment value of creatures from the real world.
View:
PDF
Citation:
PhD Thesis, Computer Science Department, The University of Texas at Austin, Austin, Texas, December 2014. Tech Report TR-15-01.
Bibtex:

Dan Lessin Ph.D. Alumni dlessin [at] cs utexas edu
Risto Miikkulainen Faculty risto [at] cs utexas edu