Intelligent Counterforce Allocation Method Using Multi-Agent Reinforcement Learning for Ground Operations

Modern military operations require commanders to make complex tactical decisions involving the effective allocation of multiple allied forces to counter enemy threats while managing diverse streams of battlefield information. This study proposes the Intelligent Counterforce Allocation for Ground Operations (CAGO) method, which leverages multi-agent reinforcement learning (MARL) to support command-level decision-making in ground warfare. Designed for brigade-level and higher force optimization, CAGO enables effective counterforce deployment in defensive, offensive, and simultaneous operations. The method models key tactical actions, such as maneuver force allocation by avenue of approach and artillery targeting, as MARL agent actions, with doctrinal tactical factors embedded into the reward function. The MARL agent is implemented using the Soft Actor-Critic algorithm and scaled through a Graph Attention Network to manage over 40 heterogeneous agents (infantry, tanks, and artillery). Compared to rule-based approaches, CAGO demonstrates lower performance variance and significantly improves doctrinal metrics, including up to a 30% increase in win rate and enhanced protection of critical allied zones, even under uncertainty and diverse operational scenarios. These results highlight the potential of MARL to enhance intelligent, large-scale force allocation under constrained human resources.