A variational approach to the optimization of gait for a bipedal robot

This paper presents a method for optimizing the walking motions of a planar five-link biped. The technique starts with non-linear kinematic and dynamic models for both the single-support and impact stages of motion. A variational technique is then used to derive joint trajectories that minimize a simple cost function. The resulting two-point boundary value problem is solved using a finite difference technique, with trajectories obtained from a simple linearized model as initial conditions. Families of optimal trajectories for different step periods and step lengths are presented.