Tracing motivation in virtual agents
Loading...
Files
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of New Brunswick
Abstract
Mathematical functions designed similar to emotional concepts in order to provide tools for
emotion-based choice are investigated. Emotions have been treated as a disturbing factor to
logical assumptions and decisions. Now they are rather treated as a necessity and a
requirement for agents in complex or not fully graspable situations with complicated or
concurrent goals. The software of SOCIAL, the 'Simulation Of Consciousness In Artificial Life'
was developed to provide a research universe for virtual agents named BUGs. The emotionbased
models of Affect Logic, Subsumption and Fungus Eater were compared against 2
randomizing benchmark approaches. To compare success of various routines, agents
equipped with decision making strategies designed similar to these theoretical approaches
were tested in predator-prey simulations with limited energy supply. The quantitative aspect
of survival was measured through experiments. Experiments include three kinds of
simulations: Simulations where all emotional routines compete inside BUGs of the same type,
simulations where all types of BUGs are equipped with the same routine, and simulations
with a mixture of type and emotional routine. The evolutionary concepts of selection and
mutation were utilized to allow the BUGs to adapt their decision making strategy to the
current simulation. For simulations run within SOCIAL, the memory based randomizing
benchmark approach outperforms all sophisticated routines. Complex models of emotional
choice seem to be not beneficial to virtual agents, but a burden instead. This is probably not
a general result but limited to the specific experiment setups tested.