Inertial Navigation with Multi-Interval Pose Measurements

The system segments the pose estimation task by creating multiple specialized machine learning models, each trained on IMU data from specific motion classes (e.g., walking, running, driving). Each model generates pose measurements independently, and the system fuses these segmented results rather than using a single comprehensive multimodal sensor fusion approach. This segmentation allows the system to achieve accurate pose tracking while consuming less energy by processing IMU data through specialized models rather than continuously activating multiple sensor modalities.

The system changes the parameter of motion class classification by training separate machine learning models for different motion classes (walking, running, driving, etc.). Each model is optimized for its specific motion type, allowing the system to maintain high reliability and accuracy for each individual motion class. The system dynamically selects or weights models based on the current motion class, ensuring reliable pose tracking across varying conditions without the reliability issues that can arise from using a single general-purpose multimodal fusion model.

The system implements universality by using multiple machine learning models that all process the same type of input data (IMU measurements) but are specialized for different motion classes. Each model is universal in its ability to process IMU data while being specialized for its motion class. The models can be selectively activated based on the current motion type, allowing the system to achieve high measurement precision across diverse conditions while managing complexity through conditional model selection rather than simultaneously running all models at full capacity.