Autonomous Navigation for Mobile Robots in Dynamic Environments: A Comprehensive Study Using Deep Learning Algorithms

Over recent decades, mobile robotics has advanced significantly, with navigation being a critical task. Robots now navigate diverse, dynamic, and unknown environments using various methods. This study reviews deep learning-based navigation techniques, highlighting their pros and cons. It focuses on a convolutional neural network (CNN)-based approach implemented in Python, C++, Robot Operating System (ROS), Linux, and the GAZEBO simulator. A notable contribution is adapting the DroNet model, originally for drones, to ground robots. However, DroNet lacks goal-oriented mobility. To address this, this paper integrates Dijkstra’s algorithm with simultaneous localization and mapping (SLAM) and path planning, enabling the robot to follow the shortest path while avoiding dynamic obstacles. Using a monocular camera instead of an expensive laser scanner, the study presents an affordable, goal-oriented navigation system suitable for indoor use in restaurants and museums.