A Robot Self-Positioning System based on Robust EKF Using Degree of Confidence

Self-positioning system is the key link of humanoid soccer robot, in which the robot needs to estimate its position and direction in the football field based on non-probing sensors. However, the modeling error of the robot system and the deviation of information acquisition bring the uncertain noise to the self-positioning system, which makes it difficult to carry out effective real-time positioning by using the traditional methods such as geometric measurement and track prediction. Existing filtering methods such as Kalman Filter and Monte Carlo Filter estimate the position and direction of the robot through observations, reducing the influence of uncertain factors. Although the accuracy of positioning is improved, the state estimation method relies on the accuracy of observations. The deviation in blurred image processing and camera ranging during the motion of the robot results in the deviation of the observations, which is more obvious in the case of multi-target observations. In this paper, a self-positioning method based on robust EKF using confidence is proposed to improve the accuracy of the robot's self-positioning by integrating observations.