Researchers have developed a new framework for Human Activity Recognition (HAR) that can handle independent orientation shifts between different body locations. This is a problem in multi-IMU settings, where inertial measurement units need to be reattached across sessions. The proposed Truly Rotation-Invariant HAR (TRI-HAR) framework uses triaxial vectors and a shared SO(3)-equivariant backbone to fuse invariant features for activity classification. According to the authors, TRI-HAR outperforms rotation-augmented baselines in four multi-IMU benchmarks. --- Why it matters: This matters because it addresses a common challenge in wearable device-based HAR, where reattaching IMUs across sessions can lead to inconsistent orientation offsets. The proposed framework provides a more robust and reliable solution for activity recognition applications. Source: https://arxiv.org/abs/2608.15621