Agentic Ai-Based Multi-Agent Reinforcement Learning for Autonomous Task Allocation and Collaborative Decision-Making

Abstract

Autonomous task allocation in multi-agent systems requires effective coordination among multiple agents with different capabilities, workloads, and operating conditions. Conventional task-allocation methods often depend on predefined rules or centralized decision mechanisms, which can limit adaptability in dynamic environments, while standard multi-agent reinforcement learning approaches may experience inefficient coordination, unstable learning, and limited high-level reasoning. This study proposes an Agentic AI-based multi-agent reinforcement learning framework for autonomous task allocation and collaborative decision-making. The framework combines task reasoning, agent capability assessment, task-agent matching, collaborative coordination, and adaptive policy learning to support efficient and context-aware decisions. A simulation-based environment is developed with heterogeneous agents, dynamically arriving tasks, varying task priorities, and changing workload conditions. The proposed framework is evaluated in terms of task allocation quality, coordination performance, learning performance, and decision efficiency using task completion rate, allocation success rate, coordination success rate, cumulative reward, convergence behavior, workload balance, communication overhead, and decision latency. The results show that the proposed method achieves a task completion rate of 94.6%, coordination success rate of 92.8%, and allocation success rate of 93.7%, while reducing decision latency by 21.4% compared with conventional MARL. These findings demonstrate the potential of Agentic AI-MARL for efficient, adaptive, and collaborative multi-agent decision-making.