Walking State Estimation for Biped Robot Using Foot Contact Information

Contact provides useful information in many applications. For bipedal walking, CoP-ZMP criterion, which is computed using the contact wrench between the foot and ground, is one of the important criteria to measure the stability of the biped robot. However, the CoP-ZMP criterion requires an accurate measure of contact point to define a support polygon, which is a difficult problem when the robot walks on uneven terrain. In this respect, this paper proposes an estimation framework, which is regardless of the support polygon, to recognize a current walking state (WS) using the contact information. The framework consists of two indicators and one discriminator. The indicators use the Gaussian Mixture Model to describe sub-states for each input. The discriminator determines the current WS by referencing rules with the outputs from the indicators. Finally, the framework is demonstrated with simulations and real robot experiments using a full-size humanoid robot, TOCABI.