Neuromorphic Digital-Twin for Multi-Agent System Control Using Spiking Neural Networks

Recent advances in multi-agent systems (MAS) have highlighted the increasing demand for safety through predictability amidst environmental uncertainties and the need for reliable autonomy. This study proposes a neuromorphic digital twin (DT) strategy for MAS, employing a simplified dynamics model on a server for virtual twins. These virtual twins evolve by learning from uncertainties and unmodeled dynamics using measurements from physical agents. Additionally, a spiking neural network (SNN) based controller is implemented at the edge in physical agents, enabling them to learn and generate signals from the virtual twins without requiring continuous communication between the virtual twins and physical twins, thereby reducing latency. Simulations validate the effectiveness of this method, demonstrating its potential for enhancing safety and performance in MAS.