Penalty kick of a humanoid robot by a neural-network-based active embedded vision system

Abstract

This paper realizes the humanoid robotic system to execute the penalty kick (PK) of the soccer game. The proposed system includes the following three subsystems: a humanoid robot (HR) with 22 degree-of-freedom, a soccer with different colors, and a soccer gate. In the beginning, the HR scans the soccer field to find the gate and the soccer, which are randomly distributed in a specific region in the front of the gate. If a command for the PK of the soccer with specific color is assigned, the HR will be navigated by an active embedded vision system (AEVS). After the HR reaches a planned position and posture, the PK of the HR will be executed. Two key important techniques are developed and integrated into the corresponding task. One is the modeling using multilayer neural network (MNN) for different view angles, the other is the visual navigation strategy for the PK of the HR. In addition, the error sensitivities in the pan and tilt directions of these four visible regions are analyzed and compared. The proposed strategy of the visual navigation includes the following two parts: (i) the switched visible regions are designed to navigate the HR to the planned position, and (ii) the posture revision of the HR in the neighborhood of the soccer in order to execute the PK. Finally, a sequence of experiments for the PK of the HR confirm the effectiveness and efficiency of the propose methodology.