Fig. 4

Evaluation of perturbation-based method. Evaluation of perturbation-based method’s ability to reproduce within-stride metabolic cost of different model-based methods in our two datasets. Estimations from each model-based method are represented by black lines. Our perturbation-based method’s estimations are represented with red lines. The left column shows evaluations in the neuromechanical simulation dataset, and the right column shows evaluations in the human experiment dataset. (A). Umberger et al., 2003 [35], (B). Houdijk et al. 2006 [33], (C). Bhargava et al., 2004 [32], (D). Lichtwark et al., 2005 [34] (E). Margaria 1968, applied onto muscle work rate [30, 52], (F). Beck et al., 2019 [31], (G). Kim and Roberts, 2015 [18] (H). Margaria 1968, applied onto COM work rate [30, 53], (I). Margaria 1968, applied onto joint work rate [30, 54] (J). Minetti and Alexander, 1997 [29]