Comparing Decentralized Ongoing Task Allocation

摘要

Real-world domains can often benefit from adaptable decentralized task allocation through emergent specialization of large groups of agents. Agents working within a shared environment can use environmental cues as a means of indirect coordination to maximize performance. We compare how groups of homogeneous agents fulfill the demands of a set of “ongoing” dynamic tasks over time, when employing Response Threshold Reinforcement, Ant Colony Optimization, or Linear Reward-Penalty Learning Automata. Results show that efficient, stable, and adaptable task allocation for large teams choosing from large sets of always-available tasks is possible, but excess and deficit workforce can present some performance challenges.