This study aims to detect hand gestures through cameras, focusing on robotics applications. Hand gestures serve as a vital mode of human-robot interaction, and enhancing the accuracy and efficiency of gesture recognition is critical for seamless communication between humans and robots. Feature pyramids play an important role in detecting objects at varying scales, and our analysis focuses on how these specific layers contribute to detection accuracy and robustness. We analyze the key roles of the P3, P4, and P5 layers in YOLO’s Feature Pyramid Network (FPN) architecture in capturing multi-scale features, essential for distinguishing subtle variations in hand gestures. Through systematic experimentation on a hand gesture dataset, we evaluate the performance of YOLO in various configurations, assessing metrics such as precision, recall, and inference time. Our findings reveal that the P5 layer demonstrated superior performance, effectively capturing intricate details of the gestures. Conversely, the P4 layer yielded the least favorable results, indicating its limitations in feature representation for this specific task. This nuanced understanding of layer contributions informs future enhancements in feature pyramid networks, with implications for improving real-time gesture recognition systems. This research aims to contribute to developing more intuitive and responsive robotic systems, facilitating natural interactions in dynamic environments.