Human behavior plays a crucial role in determining the sustainability of common-pool resources, where the misalignment of individual actions and collective goals often leads to overexploitation and resource depletion. Traditional methods, such as taxation and cost-based interventions, typically overlook cognitive mechanisms underlying individual decisions, limiting their effectiveness. We introduce the Norm-ORiented Multi-Agent Reinforcement Learning (NORMARL) framework, integrating psychologically plausible parameters—norm internalization and adaptive learning—to explore resource exploitation dynamics. Our model simulates agents interacting with a shared resource, adapting their behaviors based on feedback regarding normative consumption. Agents optimize their utility by balancing individual and social costs, influencing collective resource dynamics. We demonstrate that increased norm internalization and adaptive learning substantially enhance cooperation, resilience to environmental shocks, and resource sustainability. NORMARL highlights how incorporating cognitive mechanisms into computational models can reveal the potential of educational and behavioral interventions, complementing economic policies to promote sustainable behavior.