Distributed Adaptive Flocking Control for Large-Scale Multiagent Systems.

Area of Science:

• Robotics and Control Systems
• Multi-Agent Systems
• Game Theory

Background:

• Existing flocking methods struggle with communication complexity and the curse of dimensionality in large-scale multiagent systems (LS-MASs).
• Mean Field Games (MFG) simplify interactions using probability density functions but lack robust coordination mechanisms for efficient flocking.

Purpose of the Study:

• To develop a novel distributed flocking control method for LS-MASs in uncertain environments.
• To enhance flocking performance by addressing communication complexity and coordination challenges.

Main Methods:

• Decomposition of LS-MASs into leader-follower subgroups.
• Hybrid game theory combining cooperative, Stackelberg, and MFG for inter- and intragroup interactions.
• Hierarchical actor-critic-mass-based reinforcement learning for distributed adaptive control.

Main Results:

• A novel distributed flocking control method for LS-MASs was successfully developed.
• The hybrid game structure and reinforcement learning enabled adaptive and efficient flocking.
• Numerical simulations and Lyapunov analysis confirmed the method's effectiveness.

Conclusions:

• The proposed method effectively overcomes limitations of existing flocking techniques for LS-MASs.
• The hybrid game theory and reinforcement learning approach offers a robust solution for distributed adaptive flocking.
• This work advances the control strategies for large-scale autonomous systems in complex environments.