Multi-data fusionaided indoor localization based on continuous action space deep reinforcement learning

Abstract

Significant attention has been paid to indoor localization using smartphones in both research and industry.However, the accuracy and robustness of localization remain challenging issues, particularly in complex indoor environments.In light of the prevalent incorporation of pedestrian dead reckoning (PDR) devices in contemporary smartphones, an advanced indoor localization fusion method, anchored in the twin delayed deep deterministic policy gradient (TD3) framework, was proposed.In this approach, a seamless integration of Wi-Fi information and PDR data was achieved.The localization process of PDR was modeled as a Markov process, and a comprehensive continuous action space was introduced for the agent.To evaluate the performance of the proposed method, experiments were conducted and this approach was compared with three state-of-the-art deep Q network (DQN) based indoor localization methods.The experimental results demonstrate that the proposed method significantly reduces localization errors and enhances overall localization accuracy.

Keywords

• Wi-Fi;pedestrian dead reckoning;indoor localization;twin delayed deep deterministic policy gradient;deep reinforcement learning