Abstract
Virtual ecosystems, where natural selection is used to evolve
complex agent behavior, are often preferred to traditional
genetic algorithms because the absence of an explicitly defined
fitness allows for a less constrained evolutionary process.
However, these model ecosystems typically pre-specify
a discrete set of possible action primitives the agents can perform.
We think that this also constrains the evolutionary process
with the modellers preconceptions of what possible solutions
could be. Therefore, we propose an ecosystem model
to evolve complete agents where all higher-level behavior
results strictly from the interplay between extremely simple
components and where no ‘behavior primitives’ are defined.
On the basis of four distinct survival strategies we show that
such primitives are not necessary to evolve behavioral diversity
even in a simple and homogeneous environment
complex agent behavior, are often preferred to traditional
genetic algorithms because the absence of an explicitly defined
fitness allows for a less constrained evolutionary process.
However, these model ecosystems typically pre-specify
a discrete set of possible action primitives the agents can perform.
We think that this also constrains the evolutionary process
with the modellers preconceptions of what possible solutions
could be. Therefore, we propose an ecosystem model
to evolve complete agents where all higher-level behavior
results strictly from the interplay between extremely simple
components and where no ‘behavior primitives’ are defined.
On the basis of four distinct survival strategies we show that
such primitives are not necessary to evolve behavioral diversity
even in a simple and homogeneous environment
| Original language | English |
|---|---|
| Title of host publication | Artificial Life XI |
| Subtitle of host publication | Procs of the Eleventh International Conference on the Simulation and Synthesis of Living Systems |
| Editors | Seth Bullock, Jason Noble, Richard Watson, Mark Bedeau |
| Publisher | MIT Press |
| Pages | 474-481 |
| ISBN (Electronic) | 9780262287197 |
| Publication status | Published - 2008 |