Validation of Robustness of SLAM Algorithms using Deep Learning Methods in Real Conditions

As the level of modern technology development, namely autonomous robots, drones, robotics, etc., is high, the topic under study is highly relevant. Since the level of development of modern technologies, namely autonomous robots, drones, etc., is high, the topic under study is highly relevant. Due to the use of the Simultaneous Localisation and Mapping System (SLAM) in the industrial sector, ensuring and empirically verifying its robustness under challenging conditions is essential. The study aimed to evaluate and verify the reliability of the SLAM algorithm in real conditions. The following methods were used to conduct the study: deep learning methods and recurrent neural networks. ATE and RPE metrics were used to measure the accuracy of maps and trajectories. The study revealed a relatively high stability of the developed SLAM algorithm in changing lighting conditions and dynamic objects' presence. The ATE and RPE metrics were within acceptable limits. The study's scientific novelty and originality lie in considering the real conditions during the experiment, such as different lighting and dynamic objects, which were rarely considered in previous studies. The developed algorithm will be helpful for autonomous systems and in the context of the latest advanced technologies and robotics. A promising area for further research may be improving the SLAM algorithm for use in tough conditions.