Largest coverage network in a robot swarm using reinforcement learning.

Establishing a large adaptive connected network for decentralized swarms is useful for their behavior to share information about the working environment. A hard-coded implementation is time-consuming to achieve. Therefore, we are motivated to explore the benefits of reinforcement learning (RL) to learn a suitable adaptive policy. We also explore the combined use of a scalar field, which was inspired by template pheromones in social insects. In this paper, we investigate using RL with low and high-resolution scalar fields to solve the largest covering network problem. Our results show that RL outperforms the hard-coded approach in the presence of the high-resolution scalar field. [ABSTRACT FROM AUTHOR]