Analysis of Evolved Response Thresholds for Decentralized Dynamic Task Allocation

In this work, we investigate the application of a multi-objective genetic algorithm to problem task allocation in self-organizing, decentralized, threshold-based swarm. We use evolve response thresholds for simulated swarm engaged dynamic problems: two-dimensional and three-dimensional collective tracking. show that evolved not only outperform uniformly distributed but achieve nearly optimal performance on variety tracking instances (target paths). More importantly, demonstrate some generalize all other instances, eliminating need new each instance be solved. analyze properties allow these paths serve as universal training they are quite natural. After priori evolution, our system static. The solved by swarms highly dynamic, with schedules demands change over time significant differences rate magnitude change. That is able results refutes common assumption must perform well environment.