Multi-phase Updating - A Practical Approach to Simulating Animat Agents

摘要

A growing number of applications can be modelled using spatial agent systems or animats. Typical animat simulations model collective macroscopic phenomena using encoded individual behaviours for microscopic agents. In spatial agent systems there are well known problems that occur if agent cells are updated in a sequential order. This is known as the sweeping problem and it also occurs in other numerical simulations such as differential equation solvers and leads to observable macroscopic effects in the simulated system that are solely artifacts of the implementation algorithm. We explore various strategies to remove these artifacts including multi-phase updating. We find that in some models that include non-diffusional effects such as predator-prey interactions, a two-phase update is not sufficient and a three-phase update strategy is necessary to preserve model semantics, particularly when concurrency is used to speed up the modpl. We discuss the computational implications of this for animat agent simulations.