<p>As neuroscience develops, efforts have been made to add cognitive features to social robots. Social robots previously controlled by a user are now being trained in ways compatible with neuroscience to interact independently with humans in a social environment without the need for a user after an initial setup. When interacting with family, friends, colleagues, or strangers in different social situations, humans generally perform appropriate behavior called social etiquette. Social robots interacting with humans are also expected to perform these social etiquette rules. The purpose of this article is first to teach a robot these social manners using a video-masked autoencoder (with the vanilla ViT backbone) using modern deep learning methods, test the robot’s use of the manners, and finally evaluate if the robot can appropriately interact with humans in different social situations. For this purpose, 19 people were asked to perform 16 social customs related to body language at least five times in the learning phase. Then, the identified social manners were taught to the robot, and during the subsequent test phase, the robot performed these trained behaviors while interacting with new participants. Results showed that the robot was able to classify these social behaviors with 83% accuracy.</p>