Data-Driven H∞ Output Consensus for Heterogeneous Multiagent Systems Under Switching Topology via Reinforcement...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Robotics

Background:

• Multiagent systems face challenges with unknown dynamics, external disturbances, and dynamic network topologies.
• Leader-following tasks are complicated by missing information due to switching topologies.

Purpose of the Study:

• To develop a model-free policy gradient reinforcement learning algorithm for discrete-time heterogeneous multiagent systems.
• To address tracking problems in systems with unknown dynamics, external disturbances, and switching topologies.

Main Methods:

• A distributed adaptive observer estimates the leader's unknown dynamics and state.
• An exponential discount value function and discrete-time game algebraic Riccati equations (DTGAREs) are derived for control strategy.
• A data-based policy gradient algorithm approximates DTGARE solutions online, avoiding explicit system knowledge.

Main Results:

• The proposed algorithm successfully solves the tracking problem for discrete-time heterogeneous multiagent systems.
• The use of an offline dataset and experience replay enhances data utilization efficiency.
• Stability of the systems is ensured through exploration of the exponential discount value's lower bound.

Conclusions:

• The novel algorithm provides a robust and efficient solution for multiagent tracking under complex conditions.
• Model-free reinforcement learning combined with adaptive observers offers a powerful approach for decentralized control.
• Simulations validate the effectiveness of the proposed method in ensuring system stability and achieving accurate tracking.