A new particle filtering method for robot sensors autonomous positioning

The positioning is to estimate the position and posture of the robot and accurate position estimation is necessary to achieve autonomous navigation. So, the study of the positioning method of the mobile robot has a very important significance. Now there are three mainly approximate techniques types: the first is the probability density function on the approximate state space. Extended Kalman filter (EKF) as well as the EKF variants, approximate Bayesian formula with the increase of the traveling distance, the accumulated error will grow indefinitely. Therefore, dead reckoning method is only suitable for short-term and short-distance position estimation and with low precision. But all the particle filter algorithms is deficient in that weight variance with the random time increasing, sample weight gathers together into a small number of samples. To solve this problem, this paper presents the algorithms based on neural networks, and the importance of sample to adjust the particle filter (NNISA-PF). The technical data derives from laboratory and in-situ experimentation. The algorithm contains the establishment of robot motion system, the sensor observation model and the particle filter based on GRNN sample adjustment. Also the algorithms takes advantage of the simulation plat form provided by Alberto Vale to make a further analysis of the results of robot localization, and the conclusion suggests that through combining neural network theory with particle filter, so that the samples effectiveness and diversity to be kept and reduce sample dilution in the re-sampling stage. The experimental results show that the improvement measures can effectively improve the performance of the algorithm, so that it enables them to maintain a reliable positioning.