Multi-Agent System Cooperation Via Constructivst Learning Approach

Abstract

Multi-agent systems have a wide range of real-world applications in many fields, where multiple agents must cooperate to achieve their global objectives in a shared environment.

Cooperative learning in multi-agent systems is a fundamental field within artificial intelligence. It aims to design autonomous agents capable of cooperation by learning and adapting to dynamic environments. As these environments become more complex, agents require learning strategies that allow them not only to react but also to evolve and build their own behavioral models to solve tasks collaboratively. Constructivist approaches consider agents as active entities capable of constructing their internal knowledge through experience and intrinsic motivation, without relying on predefined behaviors or external rewards. In this work, we propose a model that integrates constructivist learning concepts like schema and intrinsic motivation in cooperative multi-agent systems to enable agents to progressively build, refine behavioral schemas and to learn collaboration behaviors without external supervision.Furthermore, we evaluate the proposed architecture for solving the collaborative resources extraction problem in a grid environment simulation. The result shows that the agents are able to learn how to navigate and how to collaborate to extract resources and avoid obstacles.The emergence of collaborative and navigational behavior through constructivist learning and intrinsic motivation mechanisms can lead to the development of autonomous, self-motivated agents capable of cooperation.