Cooperative Game-based Intelligent Actions Making for Constrained Multi-agent System

In this paper, an intelligent action-making model for constrained multi-agent systems (MAS) with cooperative games is proposed. This model aims to prevent system breakdowns triggered by the large number of agents, a prevalent dilemma in MAS across various practical situations. A widely accepted issue in the domain is that reinforcement learning methods require vast amounts of training data, which can be costly and time-intensive to acquire. Notably, existing game theory methods present challenges, including incomplete information about the agents' current strategies and the reliance on computationally intensive solutions to determine equilibrium points. To address these concerns, several novel contributions are offered in this paper. First, a new intelligent action-making model designed for large-scale MAS is introduced, ensuring effectiveness and efficiency. Second, system robustness and adaptability in intricate scenarios, especially under saturation constraints, are enhanced through the incorporation of forward prediction for more precise action-making. Additionally, a method to generate a hybrid task coalition, accounting for limited execution capabilities, is devised considering multitask constraints. This strategy aims to mitigate the lag in coalition formation due to the expansive dimensionality of action spaces in conventional methods. Simulations conducted with nine agents attest to the efficiency of the proposed model.