<p>Visual perception is crucial in robotics research, enabling robots to effectively analyze, understand, and engage with their surroundings. This research examines several aspects of visual perception, including identifying objects, interpreting scenes, navigation, and human–robot interaction, with the goal of improving robots' capabilities in real-world situations. The research explores methods designed specifically for robotic systems to identify and classify items, leveraging advances in computer vision and deep learning to help robots accurately identify objects in a variety of environments. This paper explores techniques for interpreting scenes and mapping semantics, enabling robots to form detailed descriptions of their environments and identify spatial connections. This paper proposes a CNN-VSLAM framework that combines convolutional neural networks with visual simultaneous localization and mapping methods in robotics. The algorithm helps robots understand and engage with their surroundings using visual perception. The algorithm enables robots to independently traverse complex situations, avoid collisions with objects, and improve their ability to move and adapt to changing environments. The research explores the impact of visual perception on human–robot interaction, studying gesture recognition, facial expression assessment, and other human-focused forms of visual perception. The goal is to improve intelligent robotic systems for practical use by using a variety of methods and working across different fields. This advancement improves robots' observation capabilities and expands their use in areas such as self-driving cars, warehouse operations, medical assistance and search and rescue operations, changing the landscape of robotics advancement and human–machine interaction.</p>