Human vision is remarkably robust to noise and distortions, providing a key source of inspiration for improving adversarial robustness in artificial intelligence (AI) models. Unlike standard deep networks, which are vulnerable to adversarial attacks that exploit model weaknesses to induce misclassifications, biological visual systems maintain stability under challenging conditions. Motivated by this, our study investigates two key questions to enhance AI robustness. First, we propose that human neurodevelopmental-inspired training procedures, specifically blur training and blurry-to-clear training, can enhance the robustness of VOneNet—a primary visual cortex-based architecture—against adversarial attacks, outperforming traditional training methods. Second, we examine how VOneNet, trained with these brain-inspired training regimes, compares in robustness to standard convolutional neural network architectures. Experiments on MNIST, CIFAR10, and ImageNet100 show that VOneNet, trained with these methods, achieves significantly improved robustness compared to traditional models. While both strong blur training and weak blur training enhance robustness, blurry-to-clear training provides the most substantial improvement by effectively balancing spectral bias towards low-frequency details and reducing sensitivity to adversarial perturbations. Our work highlights the value of aligning AI model training with neurodevelopmental processes, offering a promising approach to improving AI adversarial robustness.