Previous studies have used foot-ground contact models to reproduce experimental GRFs for walking [2–4,18–29], running [3,4,25,30–32], and running jumps [33] (Table 1). Most of these studies modeled foot-ground interactions using a grid of viscoelastic elements placed on the bottom of the foot, with the most notable differences between models being the number and location of elements and the number of foot segments. Researchers have placed viscoelastic elements at a single point [31], along the midline [19,20] or CoP [24] of the foot, at a small number of specific points on the bottom of each foot segment [4,18,21, 22,27,28,30,33], and at 30 or more locations under the foot [2,3,26,32]. Most studies used either a single-segment [3,22, 25,28,30,31,33] or two-segment (HF and FF) [2,4,18–21, 24,27,29] foot model. GRFs produced by single-segment foot models exhibited more discontinuities, especially at heel strike, than did those produced by two-segment foot models, possibly due to their inability to account for rolling motion under a single rigid body. In contrast, new constraint-based single-segment foot models that account for rolling motion have produced realistic GRFs without discontinuities [25]. However, the rolling constraint cannot be used in predictive optimizations because of its dependence on measured CoP data. Some two- and three-segment foot models have also produced discontinuous and/or oscillatory GRFs [18,24,26,27,29]. Results improved when the contact model was modified (e.g., viscoelastic versus volumetric) or when the number of contact elements was increased. Nonetheless, many reported models do not accurately reproduce experimental GRF curves, which is problematic for predictive gait optimization studies that require accurate prediction of GRFs to predict physically realistic walking patterns and corresponding joint moment or muscle force controls.