Algorithmic differentiation improves the computational efficiency of OpenSim-based trajectory optimization of human movement
Fig 4
Comparison of computational time (top) and number of iterations (bottom) between exact and approximated Hessian. The comparisons are expressed as ratios and averaged over results from different initial guesses (error bars represent ± one standard deviation). The horizontal lines indicate 1:1 ratios. Ratios larger than one indicate slower convergence (top) and more iterations (bottom) with an exact versus an approximated Hessian. Pend indicates pendulum simulations with the number being the number of degrees of freedom; Pred and Track indicate predictive and tracking simulations, respectively. The results were obtained using all solvers and AD-Recorder.