Distributed learning-based state prediction for multi-agent systems with reduced communication effort

A novel distributed event-triggered communication for multi-agent systems is presented. Each agent predicts its future states via an artificial neural network, where the prediction is solely based on own past states. The approach is therefore scalable with the number of agents. A communication is triggered if the discrepancy between actual and predicted state exceeds a threshold. Numerical results show that this approach reduces the communication effort remarkably compared to existing methods.