i'm curious about how effective path tracking can be in comparison with computer vision based inverse kinematics of the body itself. do all forms of bad form have detectable imu signatures?

i wonder if it would make sense to consider it as a data problem, capture a bunch of high fidelity inverse kinematics data for various forms of bad form/dangerous lifting along with the imu data and then work from there. there could be some interesting and unexpected features that are easier to detect than straying from straight line paths with some tolerance.