Wrist-worn inertial measurement units (IMUs) are widely used for tasks such as activity and gesture recognition, pose estimation, inertial odometry, and clinical motor assessment. However, comparing different sensor representations across these varied applications has been difficult due to inconsistent evaluation practices.

The new paper introduces IMU-Bench, a benchmark designed to evaluate wrist IMU representations across kinematic, semantic, and clinical tasks. By grouping tasks into these three categories, the benchmark aims to provide a common ground for assessing how well different representations generalize.

The abstract notes that existing wearable benchmarks are limited, though the full details of the proposed benchmark and its results are not included in the announcement. The paper is available as a preprint on arXiv.