Efficient neural network approach of self-localization for humanoid robot

Robot soccer game is one of the significant and interesting areas among most of the autonomous robotic researches. Following the humanoid soccer robot basic movement and strategy actions, the robot is operated in a dynamic and unpredictable contest environment and must recognize the position of itself in the field all the time. Therefore, the localization system of the soccer robot becomes the key technology to improve the performance. This work proposes efficient approaches for humanoid robot and uses one landmark to accomplish the self-localization. This localization mechanism integrates the information from the pan/tilt motors and a single camera on the robot head together with the artificial neural network technique to adaptively adjust the humanoid robot position. The neural network approach can improve the precision of the localization. The experimental results indicate that the average accuracy ratio is 88.5% under frame rate of 15 frames per second (fps), and the average error for the distance between the actual position and the measured position of the object is 6.68 cm.