Assessing a swarm-GAP based solution for the task allocation problem in dynamic scenarios

Abstract

Swarm-GAP is a heuristic that combines a swarm intelligence strategy with the generalized assignment problem (GAP) method. This approach is especially appropriate when there are agents engaged in a collaborative task, but in general, heuristics have drawbacks to optimize resource allocation. A previous work proposed the usage of three swarm-GAP variants to solve the task allocation problem among agents representing a group of Unmanned Aerial Vehicles (UAVs) aiming at the optimization of their resources usage applied in the context of static environments. However, there is a lack of empirical assessment of these algorithms in dynamic scenarios, i.e., with some attributes changing along the system execution. Such changes represent important features of real-world application scenarios, such as in military operations in which a number of non-expected events may happen, e.g., loss of members of the UAV-team or onboard sensor failure. Therefore, the contributions of this work are the performance evaluation of the original algorithms in dynamic context, and the extension of these algorithms to properly address more realistic dynamic scenarios. Considering changes in some attributes of the environment, a trade-off in terms of the quality in the mission performance and the overhead in the communication among the UAVs is explored. The empirical assessment of the original algorithms and the proposed extensions were performed by conducting independent replications in a scenario where the number of agents (UAVs) changes at runtime and adaptations occur autonomously to maintain the mission execution. The acquired results provide evidence that the proposed solution is capable of dealing with dynamic scenarios, covering the gap left by other works in the literature, and enriching the realism of applications in autonomous intelligent systems.