Performance Analysis of Positioning by Deep Reinforcement Learning in VLC-Based Empty Office Environment

Recently, buildings in cities have been becoming increasingly large, resulting in more complex indoor environments. Consequently, there is a growing demand for indoor Location-Based Services (LBS). High-precision indoor positioning technology is required to provide high-quality LBS. While relatively precise positioning can be achieved outdoors using the Global Positioning System (GPS), indoor positioning accuracy deteriorates due to walls and obstacles. Therefore, many recent studies have proposed using Radio Frequency (RF)-based wireless communication technologies, such as WiFi, Bluetooth, and Ultra-Wide Band (UWB), for indoor positioning. Among these techniques, fingerprinting is commonly adopted because it is cost-effective and can achieve high accuracy. However, RF-based technologies are prone to significant fluctuations in the Received Signal Strength Indicator (RSSI) due to environmental uncertainty. Furthermore, distinguishing between floors in a multi-story building or precisely localizing on a single floor remains challenging. To address these issues, this paper proposes a method that integrates fingerprinting with Deep Reinforcement Learning (DRL), an Artificial Intelligence (AI) technique, in an indoor Visible Light Communication (VLC) scenario. VLC utilizes light for data transmission, enabling reliable floor separation. In contrast to RF, VLC can effectively mitigate signal interference and variations. Additionally, DRL can significantly reduce the matching time required for fingerprinting. To achieve this goal, we compare and analyze the performance of DRL algorithms, including Deep Q-Network (DQN), Double DQN and Dueling DQN. Experimental results demonstrate that the proposed approach is highly efficient in terms of both processing time and positioning accuracy.