Dynamic balance of a NAO H25 Humanoid robot based on model predictive control

This paper proposes a model predictive based approach for dynamic balance of a NAO H25 Humanoid robot. Due to the inherent mechanical properties of a humanoid robot, it has a huge potential to lose its stability during its movements. Using an integrated dynamic model leads to control three potential situations a humanoid robot may encounter, namely under-actuation, over-actuation, and fully-actuation. In this paper, by resorting to the so-called Zero-Momentum Point (ZMP) criterion for robots stability, an investigation is carried out on the stability equations and movement patterns in their relevant state space regarding to choosing ZMP as a sustainability criterion, to the end of achieving an appropriate path for robot's center of mass and two legs. In order to apply the concept of the Model-Predictive Control (MPC) based walking pattern, the ZMP state space equations are distributed along the horizon length, besides defining a specific cost function, which covers control input and ZMP path. Some linear constraints are applied to ZMP implementation and robot steps, which leads to optimize the cost function. This procedure is followed by evaluating how the weight of different parts of the cost function influences its performance. Implementation and simulation in MATLAB, Python, and ADAMS estimates and verifies the MPC performance.