Hybrid Game-Theoretic and Reinforcement Learning for Adaptive Radar Jamming Decision-Making

This paper presents a novel decision-making framework for adaptive radar jamming, termed Hybrid Game-Theoretic and Reinforcement Learning (GTRL), which fuses game-theoretic (GT) priors with deep reinforcement learning (RL) through a confidence-weighted policy fusion mechanism. While RL-based jammers excel in adapting to dynamic environments, they often lack strategic foresight; conversely, GT methods offer anticipatory advantages but remain rigid in non-stationary scenarios. Hybrid GTRL bridges this gap by incorporating Stackelberg game-based initialization, utility-guided action pruning, and confidence-aware policy fusion that dynamically balances the influence of GT and RL agents. Experimental results demonstrate that Hybrid GTRL consistently outperforms both RL-only and GT-only baselines across key metrics, including average jamming utility, policy volatility, and convergence efficiency. Ablation studies confirm the non-redundancy of each architectural module. Despite moderate computational overhead, the system remains suitable for deployment on edge-grade hardware. Hybrid GTRL represents a promising step toward intelligent, resilient jamming agents capable of operating effectively in adversarial, rapidly evolving electromagnetic environments.