Output-Feedback Global Consensus of Discrete-Time Multiagent Systems Subject to Input Saturation via Q-Learning...

Area of Science:

• Control Theory
• Artificial Intelligence
• Networked Systems

Background:

• Saturated discrete-time multiagent systems (DTMASs) pose challenges for achieving global consensus.
• Previous methods often require system dynamics and network topology information for low-gain feedback (LGF) control.
• Existing approaches typically achieve only semiglobal consensus.

Purpose of the Study:

• To develop a Q-learning (QL)-based algorithm for global consensus in saturated DTMASs via output feedback.
• To overcome limitations of prior methods by not requiring knowledge of system dynamics or network topologies.
• To achieve global consensus, improving upon previous semiglobal results.

Main Methods:

• Determining the lower bound of nonzero eigenvalues for Laplacian matrices in directed network topologies.
• Defining a test control input and formulating a Q-function to derive a QL Bellman equation.
• Utilizing limited iterations of the QL algorithm to obtain the output-feedback gain (OFG) matrix.

Main Results:

• The proposed QL algorithm enables global consensus in saturated DTMASs.
• The OFG matrix is derived without needing agent dynamics or network topology information.
• The algorithm demonstrates effectiveness through two simulation examples.

Conclusions:

• The QL-based output feedback control is effective for achieving global consensus in saturated DTMASs.
• This approach offers a significant advancement by eliminating the need for detailed system and network information.
• The method provides a robust solution for consensus problems in complex multiagent systems.