Enhancing ZMP-based Control Model: System Identification Approach

The approximation of humanoid robot by an inverted pendulum is one of the most used models to generate a stable walking patterns using a planned Zero Moment Point (ZMP) trajectory. However, on account of the difference between the multi-body model of the humanoid robot and the simple inverted pendulum model, the ZMP error might be bigger than the polygon of support and the robot falls down. To overcome this limitation, we propose to improve the accuracy of inverted pendulum model using system identification techniques. The candidate model is a quadratic in the state space representation. To identify this system, we propose an identification method which is the result of the comprehensive application of system identification to dynamic systems. Based on the quadratic system, we propose as well a controlling algorithms for on-line and off-line walking patterns generation for humanoid robots. The efficiency of the quadratic system and the walking patterns generation methods has been successfully shown using dynamical simulation and conducting real experiments on the cybernetic human HRP-4C