Distributed and Decentralized Control and Task Allocation for Flexible Swarms

TL;DR

A comprehensive model comprised of agent, formation, and swarm layers is proposed in this paper, where each layer performs a specific function in shaping the swarm's collective behavior, thereby contributing to the emergence of the anticipated behaviors.

Abstract

This paper introduces a novel bio-mimetic approach for distributed control of robotic swarms, inspired by the collective behaviors of swarms in nature such as schools of fish and flocks of birds. The agents are assumed to have limited sensory perception, lack memory, be Identical, anonymous, and operate without interagent explicit communication. Despite these limitations, we demonstrate that collaborative exploration and task allocation can be executed by applying simple local rules of interactions between the agents. A comprehensive model comprised of agent, formation, and swarm layers is proposed in this paper, where each layer performs a specific function in shaping the swarm's collective behavior, thereby contributing to the emergence of the anticipated behaviors. We consider four principles combined in the design of the distributed control process: Cohesiveness, Flexibility, Attraction-Repulsion, and Peristaltic Motion. We design the control algorithms as reactive behaviour that enables the swarm to maintain connectivity, adapt to dynamic environments, spread out and cover a region with a size determined by the number of agents, and respond to various local task requirements. We explore some simple broadcast control-based steering methods, that result in inducing "anonymous ad-hoc leaders" among the agents, capable of guiding the swarm towards yet unexplored regions with further tasks. Our analysis is complemented by simulations, validating the efficacy of our algorithms. The experiments with various scenarios showcase the swarm`s capability to self-organize and perform tasks effectively under the proposed framework. The possible implementations include domains that necessitate emergent coordination and control in multi-agent systems, without the need for advanced individual abilities or direct communication.

Figures (10)

• Figure 1: A school of fish: the school formation changes to increase survivability, yet keeps cohesiveness, in response to the predator's presence. Photograph by the first author, taken in the Maldives in $2022$.
• Figure 2: Description of the swarm multi-layered behavior that emerges from the local dynamics rules
• Figure 4: Two agents $a_i,a_j$ repel each other when mutually visible. The left image depicts the agents' movement direction within $AR_{i,j}$ as a result of this repulsion. The right image illustrates the agents' locations after some time, where they are positioned on the boundary of their $AR$ and can maneuver only on their $AR's$ circumference.
• Figure 6: Adding random walk steps for simulating the peristaltic effect within the agents' $AR$.
• Figure 7: The influence of flexibility on the swarm size over time (from left to right). The figure shows two rows of images: the top row displays swarm exploration under the "Never Lose a Friend" algorithm, while the bottom row illustrates exploration under the "Flexible Swarms" algorithm. The comparative bounding radius graph illustrates the difference between the two algorithms throughout the entire scenario.