Inertial Pose Estimation Method Based on Multi-Genre Cascade Networks

Attitude estimation using an inertial measurement unit is a critical challenge in aerospace, robotics control, unmanned systems, and related fields, with its precision directly impacting the performance and safety of navigation systems. Conventional methods for attitude estimation typically rely on intricate mathematical models and human expertise, and their accuracy is constrained by the intricate noise characteristics of sensors like accelerometers and gyroscopes. In response to this issue, the multi-genre cascade networks methodology is introduced, which integrates recurrent neural networks, multiple attention layers, and echo state networks to enable feature extraction and modeling of inertial measurement data. Experimental validation using publicly available datasets demonstrates that the fused results exhibit superior accuracy compared to conventional fusion methodologies and single neural network fusion strategies.