Development of An Intelligent Mobile Robot System for Object and Obstacle Detection

Abstract

This study presents the development of an intelligent autonomous mobile robot system for object detection and classification using the combination of Faster Regional-Convolutional Neural Network (R-CNN) with a Double Deep Q-Learning Network (DDQN) for decision-making in a dynamic environment. The system adopts the Open ImageV5 data-set for testing and training, improving model performance through the use of strong preprocessing and augmentation methods. While the DDQN allows for optimum path planning and manoeuvring decisions in stochastic mobility circumstances, the faster R-CNN guarantees excellent detection accuracy by improving object categorisation and bounding box predictions. The suggested concept was put into practice and assessed in a gaming environment that mimics actual circumstances. The results presents a consistent performance in braking and stopping distances, a processing speed of 5–10 frames per second, and good detection accuracy (90–95%). The system's generalisability and dependability were confirmed using a 5-fold cross-validation approach, which also confirmed that it is appropriate for real-time applications. With applications in robotic systems, security, and environmental monitoring, this study demonstrates the possibility of combining deep learning with reinforcement learning for autonomous navigation.