Implementation of Neural Networks in Real-Time Swarm Robotics Applications

Over the last few decades, significant attention has been devoted to the coordinated motion of swarm systems, which involve multiple autonomous robots. These systems have a range of potential applications, such as creating formations with physical or non-physical bonds, navigating to desired locations while maintaining formation, and preventing collisions. To achieve coordinated motion, recent researches have focused on the potential function method. We previously validated novel potential functions and developed a sliding-mode speed controllers to promote collective behavior. In this study, we implement an accepted neural network with one hidden and one output layer to update potential function weights with decentralized manner in real-time, thereby enhancing swarm behavior for formation creation and maintenance. The effectiveness of this approach is demonstrated through simulation and processor-in-the-loop tests using two different micro-controllers as well as measuring execution time of neural networks' models. By successfully validating neural networks implementation in this context, our study extends the current understanding of swarm behavior and opens up new possibilities for real-time swarm applications.