Overcoming the Intrinsic Performance Limitations of MEMS IMU via Diffusion-Based Generative Learning

In three linesA conditional diffusion model based on U-Net architecture synthesizes high-fidelity virtual IMU data from low-cost IMU measurements. Trained with high-grade IMU measurements as ground-truth priors, the model significantly improves positioning and attitude estimation accuracy, and produces thinner, more consistent point clouds in airborne mapping experiments.

Summary generated by Claude — human-verified