Predictably un-predictable — on the implementation of a Walking Pattern Generator for the full-sized humanoid robot HUBO2 using Model Predictive Control

There is a large body of research related to WPG (Walking Pattern Generators) for humanoid robots. Typically WPG are evaluated based on how well the robot’s actual ZMP (Zero Moment Point) tracks the desired ZMP trajectory, using a simulation of a rigid-body robot walking on a solid floor. However little has been written about how various approaches scale-up to a full-sized humanoid robot, which has unmodeled compliance in the joints or contact surfaces, and makes contact with non-rigid surfaces in the environment like a soft floor. This paper compares the implementation of three WPG: Parametrized Polynomials, Preview Control and MPC (Model Predictive Control), and shows results from simulation and initial testing on the 1.25m tall humanoid robot HUBO2.