Cross-scenario Adaptive Indoor Positioning Based on Meta-learning and Multimodal Gradient Fusion

In view of the large data distribution differences, insufficient adaptability to small samples and multimodal gradient conflicts faced by traditional indoor positioning methods in cross-scenario applications, this paper introduces an adaptive positioning method based on meta-learning and multimodal gradient fusion. The MAML framework is used to extract common features across scenarios, and the attention mechanism is combined to dynamically allocate the gradient weights of Wi-Fi (Wireless Fidelity), Bluetooth (BT) and inertial sensors to optimize the efficiency of multimodal data fusion. After fine-tuning this method in the new scenario with a small amount of labeled data, the root mean square error (RMSE) of positioning is reduced to 0.82 meters, and the cross-scenario transfer success rate (TSR) reaches 92%, which significantly improves the positioning accuracy and generalization ability, and provides an effective solution for adaptive positioning in complex indoor environments.