A Lyapunov-based MPC for Distributed Multi Agent Systems with Time Delays and Packet Dropouts using Hidden Markov Models

TL;DR

This paper introduces a novel Lyapunov-based Model Predictive Control framework that uses Semi Continuous Hidden Markov Models to handle network imperfections like delays and packet dropouts in distributed multi-agent systems, with online learning capabilities.

Contribution

It combines SCHMM with Lyapunov MPC to adaptively mitigate network and topology errors in multi-agent systems under communication uncertainties.

Findings

Effective in reducing control effort and network-induced errors.

Maintains optimality despite network imperfections.

Demonstrated adaptability across various topologies.

Abstract

We propose a SCHMM LMPC framework, integrating Semi Continuous Hidden Markov Models with Lyapunov based Model Predictive Control, for distributed optimal control of multi agent systems under network imperfections. The SCHMM captures the stochastic network behavior in real time, while LMPC ensures consensus and optimality via Linear Matrix Inequalities LMIs. The developed optimal control problem simultaneously minimizes three elements. First, the control effort is reduced to avoid aggressive inputs and second, the network induced error caused by time delays and packet dropouts. Third, the topology-induced error, as the distributed graph restricts agents access to global information. This error is inherent to the communication graph and cannot be addressed through offline learning. To overcome this, the study also introduces the incremental Expectation Maximization EM algorithm, enabling online learning of the SCHMM. This adaptation allows the framework to mitigate both network and topology errors while maintaining optimality through MPC. Simulations validate the effectiveness of the proposed SCHMM LMPC, demonstrating adaptability in multi agent systems with diverse topologies.