Inverse dynamics is commonly applied to the biomechanical analysis of human movement in order to estimate the net muscular torques generated in various limb joints. Past research has shown that errors in these torque estimates can be relatively large. These uncertainties are attributed to inaccuracies in the input variables of the inverse dynamics equations. A preliminary study to this dissertation suggested that the two main contributors to these uncertainties are inaccuracies in the estimated segment angle profiles, and inaccuracies in the estimated anthropometric parameters used to define the body segments.

To improve the accuracy of inverse dynamics estimations, it is necessary to find techniques to reduce the effects of these error sources. One method for reducing the error relies on the principle that, in an ideal case without errors, the ground reaction forces applied to the feet (GRF) as calculated using a top-down, inverse dynamics approach should match the actual GRF measured from a force plate(s) under the feet. This principle was used to formulate an optimization problem with a cost function that minimized the difference between the measured and calculated GRF. Past optimization studies were limited as their performance deteriorated when multiple sources of error were present (e.g., errors in measured motion and body segment parameters). To address this limitation, this dissertation proposed the development of an optimization-based approach that can accommodate errors due to both. The development of this method included several studies. Study 1 optimized for errors in measured motion using simulated data and the proposed optimization approach. Compared to values computed using traditional approaches, the optimization approach reduced the average root mean square error (RMSE) in the joint torques by 54%-79%. Study 2 expanded the method to account for inaccuracies in estimates of body segment parameters. This was achieved using a three-step sequential optimization. The average reduction across all joints was 48%. Study 3 demonstrated that the proposed method was successful with actual experimental data and identified areas for future refinement.

This research suggests that the sequential optimized inverse dynamics approach described in this research has the potential to significantly reduce the errors in joint torque calculations.