“…The advantage of this approach is that it allows us to extract data from sensors that are not directly measured by them. For instance, using machine learning models applied to sparse IMUs, we can extract 3D ground reaction forces (Ancillao et al, 2018;Revi et al, 2020;Wouda et al, 2018a), full body joint kinematics (Weygers et al, 2020;Wouda et al, 2018a), CoM velocity (Sabatini and Mannini, 2016), Centre of Pressure (CoP) trajectory (Podobnik et al, 2020) etc. A drawback of this method is that the machine model cannot be related to an underlying physical model of motion.…”