Utilizing a deep neural network for robot semantic classification in indoor environments.

The utilization of semantic knowledge has ushered in a new era in robot navigation and localization, enabling heightened information representation. This paper introduces an enhanced semantic classification system that leverages a cost-effective, low-processing LiDAR unit in conjunction with a proficient deep neural network (DNN) model. Unlike vision-based methods, which are often susceptible to lighting conditions and environmental variability, LiDAR offers more robust and consistent performance in diverse settings. The Robot Operating System (ROS) development environment was employed alongside a two-wheel-drive robot platform to evaluate the system's efficiency and accuracy. The efficacy of the proposed system has been rigorously validated through both simulation studies and real-world scenarios across two distinct experimental testbeds characterized by varying features. Encouragingly, the results obtained showcase a high level of semantic classification accuracy, standing competitively against diverse semantic classification systems. Furthermore, the developed system successfully generated a semantic map of the navigational area with exceptional classification precision.