Autonomous Robot Navigation System Workflow for Monitoring and Maintenance in Industry 4.0 Applications

Autonomous navigation in factories faces a different challenge with a lack of GPS, frequently changing environment, and human interference. Current methods employed include autonomous guided vehicles, which require extensive setup and lack flexibility, making it unsuitable for frequently changing environments. Prohibiting the adoption of an autonomous mobile robot is the slow mapping time and cost. A method of autonomous navigation combining computer vision with path-planning algorithms is presented. This method uses cameras attached to the environment for navigation and not on the robot to leverage security cameras commonly available. Out of the four aspects of navigation, only three were successful, namely perception, localisation, and cognition. Mapping via grayscale thresholding is found faster than simultaneous localisation and mapping but is less accurate because it is dependent on lighting. Faster region-based convolutional neural network or You Only Look Once version 4 (YOLOv4) found no difference because the travel time is significantly longer than the processing time of both. The proposed method using A* path planning with Euclidean heuristic successfully reaches the goal for at least 30 repeats. Navigational abilities are still limited in real-world settings because of accumulation of error from odometry and inertial measurement unit sensors and lack of localisation feedback.