Dynamic optimization of multi-layered defenses inspired by Chakravyuh

As adversaries grow more sophisticated, critical infrastructure demands defense systems that not only react but also learn and adapt dynamically. This work introduces a novel reinforcement learning framework inspired by the ancient Indian Chakravyuh formation, integrating Q-learning, Markov decision processes, and network optimization to model multi-layered security under uncertainty. The system enables attackers to attempt sequential node breaches while defenders deploy adaptive traps and allocate resources through quantifiable metrics including ROI-driven investment and critical node vulnerability analysis. Results demonstrate both vulnerabilities and strengths: universal Layer 0 breaches occur (Mean Time to Breach = 52 episodes) due to uneven resource allocation quantified by a high Gini coefficient of 0.712. Despite this vulnerability, deeper layers remain highly resilient — with over 90% of attacks halted by Layer 1 and fewer than 5% of episodes resulting in breaches beyond Layer 2. Trap deployment achieves high efficiency, with approximately 82% of traps being triggered, especially during early episodes. However, efficiency declines over time as attackers adapt and avoid traps. Resource allocation patterns scale linearly, ensuring sustainable defense operations. These findings validate how the fusion of Chakravyuh strategy with modern reinforcement learning creates an adaptive defense system, simultaneously exposing perimeter vulnerabilities for targeted reinforcement and demonstrating effective deeper-layer security through optimized stochastic policies.