Previous research using evolutionary computation in Multi-Agent Systemsindicates that assigning fitness based on team vs.\ individual behavior has astrong impact on the ability of evolved teams of artificial agents to exhibitteamwork in challenging tasks. However, such research only made use ofsingle-objective evolution. In contrast, when a multiobjective evolutionaryalgorithm is used, populations can be subject to individual-level objectives,team-level objectives, or combinations of the two. This paper explores theperformance of cooperatively coevolved teams of agents controlled by artificialneural networks subject to these types of objectives. Specifically, predatoragents are evolved to capture scripted prey agents in a torus-shaped gridworld. Because of the tension between individual and team behaviors, multiplemodes of behavior can be useful, and thus the effect of modular neural networksis also explored. Results demonstrate that fitness rewarding individualbehavior is superior to fitness rewarding team behavior, despite being appliedto a cooperative task. However, the use of networks with multiple modulesallows predators to discover intelligent behavior, regardless of which type ofobjectives are used.
展开▼
