An Optimization-Based Methodology to Predict Digital Human Gait Motion

New methods for fast, adaptive motion prediction of a virtual human are proposed and tested. An optimal locomotion for gait-driven motions like pushing, climbing and pick-up/delivery are sought through gradient-based optimization and inverse-dynamics. Such gait-driven motion can be produced by adapting the normal gait motion to the case when a characteristic force is applied, which is called an applied force. The applied force is a resistance force for pushing case and an object weight for delivery case. The concept of the zero moment point is modified to assess the dynamic equilibrium of the motion in presence of the applied force. For fast calculation, analytical forms of the cost/constraint gradients are provided. Stepping patterns are specified a priori to ensure the continuity of the cost/constraint function gradients. Also, by varying knots for the B-spline curve approximation, the gait stage durations are optimized.