Methodologies enabling Mobile Robots Collective Motion: A Comprehensive Review

The demand for multi-robotic systems continues to grow. As a result, there is a notable interest among researchers and industry experts to develop control methods or policies to determine how multi-robot system members should cooperate in order to enforce cohesion. These methods can be classified as nature-inspired, self-propelled particles (SPP) based, and learning-based algorithms. This paper presents a comprehensive review of these methods. The main aim is to identify, analyze and discuss the strengths and weaknesses of these methods. Additionally, this paper aims to suggest the optimal control method for enabling effective collective motion of multiple robots and to highlight the identified research gaps and suggest how they can be addressed. Nature inspired algorithms such as ant colony optimization (ACO) are simple and easy to implement when compared to others. However, they require careful parameter tuning for them to operate optimally. On the other hand, self-propelled particles (SPP) based algorithms are decentralized, easy to configure, and produce naturalistic emergent behavior, but suffer from high oscillation when experiencing inaccurate sensor readings and communication delays. Addressing the limitations of nature-inspired and SPP based algorithms should focus on developing control methods with learning abilities. Although learning-based methods are computationally intensive, they are capable of handling sensor inaccuracies and communication latency, making them well suited for the collective motion requirements of mobile robots, particularly in highly dynamic environments. Various strategies, including network pruning, the use of TinyML, and Central Training with Distributed Execution (CTDE), can be employed to optimize learning based methods for robots with limited resources.