Communication optimization for efficient dynamic task allocation in swarm robotics.

Cooperation is a central idea to the usage of swarm robotics. It enables the solution of complex problems with a coordinated execution of simple tasks by a large group of small robot, which together lead to the achievement of the swarm common goal. This coordination is only possible with an efficient task allocation. Inspired by the strategy of the particle swarm optimization algorithm, we propose a novel algorithm called the Clustered Dynamic Task Allocation. This algorithm performs task allocation in a swarm robotic system in a fully distributed manner. It performs a guided search of the allocation space using the concept of adaptive speed. This search process requires information exchange between robots. This robot communication process must be planned carefully so as to achieve two conflicting objectives: the knowledge acquired by a given robot must flow throughout the swarm so that the optimization process may converge yet this communication must be limited so it does not hinder the efficiency of the task allocation process regarding large swarms. This paper proposes the use of a clustered communication topology between the swarm robots, aiming to optimize the underlying communication process, and thus enabling efficient task allocation for large robotic swarms. The results obtained with the cluster-based topology are compared to those obtained with the full mesh-based topology. On average, the results show a clear improvement in terms of execution time and battery charge requirements. Moreover, the performance of the proposed algorithm and the stability of the produced allocation are compared to other well-known models, demonstrating its better applicability to real-world swarm robotics based systems. • We propose a novel approach to the Dynamic Task Allocation problem for Swarm Robotic based system. • The proposal is inspired by Particle Swarm Optimization. • We optimize the communication topology using a clustered configuration to reduce both execution time and battery charge usage. • We achieve a significant improvement both in terms of convergence time and battery charge requirement over the approach using a mesh topology. • We demonstrate that the proposed algorithm has real chances to be used in real-word applications based on swarm robotics when compared to existing well-known models. [ABSTRACT FROM AUTHOR]