A Convex Optimization Approach for Online Human Walking Motion Adaptation

For human-like locomotion, applying human motion capture data (HMCD) to the humanoid robot is a nature way. However, adapting HMCD online is a non-trivial problem. Not only because achieving both kinematic similarity and dynamic stability is difficult but also because such an optimization problem is intractable. This paper proposes an online adaptation approach using convex optimization for humanoid's walking motion, where the optimal objectives of kinematic similarity and dynamic stability can be simultaneously considered and easily balanced. With an inverted pendulum model based dynamics formulation of zero moment point (ZMP) in the framework of Lie group, the optimal objectives and constraints on both kinematics and dynamics are formulized into a second order cone programming problem. The simulation experiments show that the proposed method is promising for adapting HMCD online as well as simultaneously achieving and balancing the kinematic similarity and dynamic stability.