Proactive Decentralized Multi-Agent Data-Driven Node Relocation for Jammed Mobile Networks

With the rising popularity of decentralized wireless networks, their vulnerability to cyber attacks has become a pressing concern. The availability of Mobile Ad Hoc Networks (MANETs) is critical for applications like disaster recovery and emergency response, making them especially vulnerable to physical-layer threats such as jamming, where adversaries transmit high-power noise to disrupt communication. We consider critical scenarios in which MANET nodes serve as a backbone for mobile users. Existing jamming defenses often rely on non-adaptive optimization, assume simplified environments, target secondary metrics, or ignore mobility. We propose Proactive Data-driven Relocation (PADRE), a proactive decentralized Multi-Agent Reinforcement Learning-based node relocation scheme aimed at maximizing throughput during jamming. Simulations show that PADRE outperforms our proposed heuristic baseline by up to 278% in static and 585% in dynamic scenarios with realistic mobility, varying node/jammer counts and strengths. It also shows quick recovery and maintains high performance under severe signaling disruption, with nearly unimpaired results even beyond 75% signaling packet loss.