Costs and Benefits of Behavioral Specialization

Abstract

Abstract. In this work, we study behavioral specialization in a swarm of autonomous robots. In the studied swarm, a robot working repeatedly on the same type of task improves in task performance due to learning. Robots may exploit this positive effect of learning by selecting with higher probability the tasks on which they have improved their performance. However, even though the exploitation of such performanceimproving effects is clearly a benefit, specialization also entails certain costs. Using a task allocation strategy that allows the robots to behaviorally specialize, we study the trade-off between costs and benefits in simulation experiments. Additionally, we give a perspective on the impact of this trade-off in systems that use specialization.

On the design of generalist strategies for swarms of simulated robots engaged in a task-allocation scenario

“…Moreover, in some studies individual mechanisms allow single agents to act either as generalist or as specialist in response to the environmental conditions in which the group is required to operate (e.g., see Ferrante et al, 2015). Other studies investigate scenarios in which task allocation is assumed to be beneficial, either because the experimental scenario features concurrent tasks (see Ducatelle et al, 2009;Brutschy et al, 2012), as in this study, or because a group task is organised in sub-tasks that have to be executed in a predefined order (see Brutschy et al, 2014). In these studies, the focus is generally on how the competencies and characteristics of the single agents bear upon the group performance.…”

Costs and benefits of behavioral specialization

“…We evaluate the adaptivity by using the metrics F and P , as introduced in Section 4. For both measures we give the 25%,50% and 75% quantiles, as they are not normally distributed (see on-line supplementary material [4]). Fig.…”

“…Figure 6(left) reports the mean of the number of completed tasks performed at the end of the experiment, using the selective strategy (white surface) and the greedy strategy (dark surface), for different values of search speed and minimal task completion time. The number of completed tasks is normally distributed with a SE < 2%; we therefore report only the observed mean (see on-line supplementary material [4]). Fig.…”

Task Allocation in Robotic Swarms: Explicit Communication Based Approaches

“…Most solutions in self-organized task allocation is threshold-based that are inspired by models initially proposed to describe the behavior of insect societies [1]. In this case we can mention Krieger and arXiv:1503.00237v1 [cs.MA] 1 Mar 2015 Billeter work [9] which benefits from a simple threshold-based model for task allocation in a foraging scenario.Labella et al [10] and Lui et al [11,12] proposed two probabilistic task allocation approaches which use adaptive thresholds.Brutschy et al in [2] presented a task allocation strategy in which robots specialize to perform tasks in the environment in a self-organized. Jones and Mataric [8] introduced an adaptive distributed autonomous task allocation method for identical robots.…”