Aim of this project is to introduce machine learning methods to learn about the surroundings for a robot’s navigation, we want to develop an agent which will learn all the corner cases and conditions which it needs, to properly navigate, without stating some predefined rules. The whole process of learning will be carried out by using the data it will get fed in such surroundings during real life scenarios. This project will be an expansion of the previous work “Learning socially acceptable behavior using machine learning techniques on graph data” (https://github.com/robocomp/sngnn) in that project algorithms were developed to produce a single score for the robot to navigate, but it required high number of queries to work. This time we are aiming for generating these scores as a heatmap(such as the heat maps shown in https://ljmanso.com/sngnn) which can produce all the scores at once, this will be more efficient and faster than the previous work done. To solve this issue of generating all the scores at once we will try to generate a bitmap image using CNNs. As we are dealing with graphs of scenarios here we will be using GNNs.