Quadratic encoding of optimized humanoid walking

In this paper we show that optimal stepping trajectories and trajectory cost for a walking biped robot on rough terrain can be encoded as simple quadratic functions of initial state and footstep sequence. In order to find this encoding, we build a database of optimal walking trajectories for a 3D humanoid model by sampling the input space (initial state and footstep sequence) and solving a physically-based trajectory optimization problem for each sample. Then, the function coefficients are obtained by fitting the data using least squares. The performance of the proposed method is evaluated by comparing the function values with other optimal walking motion data generated with different footstep samples. As an application, we use a quadratic function to calculate the effort cost used in finding an optimal footstep sequence with an A* algorithm. Our study shows that a simple function can encode optimal walking effectively, which provides a fast alternative to online optimization of walking with full body dynamics.