Multimodal inverse kinematics significantly improves IMU-based biomechanical analyses

In musculoskeletal simulations, IMU-based approaches are often compromised by errors such as joint angle drift and offset errors due to calibration inaccuracies. These errors can compromise the accuracy of both kinematic and dynamic outcomes. This study presents a simulation-based investigation that uses synthetic inertial and positional data to systematically assess the potential of integrating spatial reference information into IMU-driven inverse kinematics analyses. Optical motion capture data was captured and error-free kinematic and dynamic data was created based on the optical motion capture data. The error-free data was then used as a reference. Based on this reference data, synthetic orientation and position data was created, incorporating a range of error types and magnitudes (e.g., sensor noise, drift, misalignment). To create the IMU-based analysis results, we calculated relative quaternions based on the orientation data which were then converted into Euler angles. We then conducted a sensitivity analysis to determine the spatial accuracy required in the position data to effectively compensate for typical IMU errors. Across all error types and magnitudes, the multimodal inverse approach (using both synthetic IMU and positional data) yielded significantly more accurate results than solely IMU-based analyses. Specifically, the mean joint angle RMSE decreased by [Formula: see text], the mean joint torque RMSE by [Formula: see text], the mean residual force RMSE by [Formula: see text], and the mean residual torque RMSE by [Formula: see text]. Future research will evaluate the effectiveness of the multimodal inverse kinematics approach when applied to real-world measurement data.