Accurate and Steady Inertial Pose Estimation through Sequence Structure Learning and Modulation

However, empirically, we find that, directly using the native transformer to model IMU sequences results in unacceptable jitter and inaccurate postures.