This study presents a comprehensive analysis of the robustness of deep learning-based gait recognition systems in the face of adversarial attacks. As gait recognition technologies become increasingly vital in security-critical domains, they are challenged by the emerging threats of adversarial interventions. In response, our research proposes a novel approach, integrating the strengths of Proximal Policy Optimization (PPO) and Generative Adversarial Networks (GANs), to engineer and analyze complex adversarial attacks. The focus of our strategy is the generation and deployment of adversarial patches, designed to disrupt gait recognition algorithms while remaining imperceptible to human observers. Utilizing reinforcement learning principles, our method strategically positions these patches, compelling the target Convolutional Neural Network (CNN) models into erroneous gait pattern classification. The effectiveness of our methodology is demonstrated through comprehensive evaluations using the CASIA Gait Database: Dataset B, a prominent dataset in gait recognition research. The results underscore a noticeable decline in the accuracy of gait recognition systems post-attack, affirming the effectiveness of our adversarial tactics.