Intrinsically Motivated Agent Behavior in a Swarm

Intrinsically motivated artificial agents are capable of open-ended exploration and cumulative learning. Swarm Intelligence is the intelligent behavior demonstrated by a group of simple agents that can solve complex problems as a group. There is relatively little work examining intrinsically motivated agents in a swarm setting. It includes a lack of metrics to measure the effect of intrinsic motivation in a swarm. This paper presents a model for a flock of agents capable of novelty detection and taking action to maximize immediate novelty. The intrinsically motivated behavior of these agents is examined in a simulated gallery environment. We also introduce behavior metrics to quantify the motivated behavior. Our results demonstrate the effectiveness of these metrics to determine the effects of the motivation mechanism as well as the exploratory behavior induced by it.