Using Different Deep Learning Models for Robot Navigation in Maze

One of the autonomous navigation tasks in a robot or self-driving car is reaching a specified location based on a series of decisions. In recent years, different path-planning algorithms have been developed for navigation and avoiding obstacles. However, all these algorithms depend strongly on the data that comes from the environment through different sensors. When the environment becomes more complex, the amount of data required to process increases, and thus the algorithms become more complex. This work implements an autonomous navigation system using only one input resource. The input resource is a video camera, and the aim is to enable a mobile robot to solve a maze. Different deep learning algorithms have been used to classify the environment state into three categories. Each category represents an action or move step for the robot. Experiments show that the approaches used work efficiently and can solve the maze. So, it can be used in some areas where robots need to solve tasks.