A LIDAR-GUIDED TRANSFORMER FRAMEWORK FOR LOW-COST BLUETOOTH AOA POSITIONING WITH ADAPTIVE KALMAN FILTERING

Bluetooth Angle-of-Arrival (AoA) positioning using deep learning has shown strong potential, but remains challenging to deploy reliably in complex industrial environments due to signal interference, dynamic obstacles, and the prohibitive cost of acquiring labeled data. To address these challenges, we propose LiDAR-guided Transformer framework for low-cost Bluetooth AoA positioning with adaptive kalman filtering (LITAK). Leveraging ubiquitous and low-cost BLE hardware, the system employs LiDAR solely during the training phase for automatic ground truth generation, eliminating the need for costly and labor-intensive manual labeling. A computationally efficient Transformer network learns the nonlinear mapping between raw In-phase and Quadrature (I/Q) samples and signal angles, which are then combined across base stations using Least Squares triangulation to estimate position. Finally, adaptive Kalman filtering ensures smooth and temporally stable trajectory output without requiring additional expensive sensors during deployment. Experiments conducted in a cluttered office environment simulating industrial conditions demonstrate a mean positioning error of 0.695 meters, with comparative analysis confirming the superiority of LITAK over standard deep learning methods.