Bayesian vs. least-squares inverse kinematics

Two simulation experiments are presented to gauge the accuracy of a new inverse kinematics method based on Bayesian inference (BIK; Pataky et al., 2019) in more realistic models than were considered previously. The first application concerns planar kinematics in the presence of soft-tissue artefacts and the second application concerns rigid body kinematics in 3D with finite helical axes (FHA). The percentage of simulations in which BIK was more accurate than least-squares based methods was only high in cases of relatively large noise magnitudes (noise SD >5 mm) or when the rotation magnitude was very small (⩽5 deg) in the 3D FHA model. Correlated parameters are the likely culprit of the low performance of BIK. Also computation time is a major deficit of the BIK approach (±20 s for the movement between two time frames). These results indicate that more research will be necessary to improve the accuracy of BIK for complex biomechanical models at realistic noise levels and to reduce computation time.