Research and Application of CNN-LSTM Fingerprint Localization for Indoor Mobile Robots in Big Data Based on CSI Signals

With advancements in mobile robot technology in China, these robots have rapidly integrated into production and daily life scenarios. Indoor positioning for mobile robots demands high accuracy and strong robustness; however, existing robots Simultaneous Localization and Mapping (SLAM) predominantly rely on Adaptive Monte Carlo Localization (AMCL), which suffers from three disadvantages: poor adaptability to dynamic environments, strong sensor dependence, and significant computational overhead. Fingerprint positioning can effectively solve these problems through wireless signal feature matching and environmental perception. Current fingerprint algorithms face several drawbacks: Received Signal Strength Indicator (RSSI) is susceptible to multipath interference; building offline fingerprint libraries incurs heigh labor costs; accuracy degradation in dynamic environments; storing and processing massive fingerprint data is challenging; and temporal features are underutilized. To thoroughly solve these problems, this paper proposes an indoor mobile robot fingerprint localization method based on Channel State Information (CSI) signals and CNNLSTM models with a big data framework. The core technologies of this method include: (1) An innovative CNN-LSTM fingerprint matching algorithm fused with AMCL; (2) Spark-driven parallel processing of massive fingerprint data. Experimental results across different scenarios confirm that the proposed method achieves a 90th percentile error of 0.65 m and exhibits strong adaptability to complex environments. This method is applicable to scenarios such as warehouse inspection robots, indoor medical service robots, etc.