Multi-sensor Fused Localization Algorithm Based on Optimized Nearest Neighbor Search

Abstract

Fast and accurate localization of mobile robot is one of the most critical problems in its field, by which the robot can obtain information about its environment and track its poses. An optimized adaptive Monte Carlo localization (BBF_AMCL) based on a 2D raster map is proposed to solve the robot localization problem. The improved localization algorithm is based on 2D laser information, IMU, odometer, and cliff sensors for localization. First, the efficient nearest neighbor search technique is used to construct the likelihood domain map, which reduces the cost time to build the likelihood domain map and speeds up particle weight updating. Second, an expanded Kalman filter is utilized to combine data from the IMU and the odometer in order to increase a priori positioning accuracy, as well as to reduce the robot's judgment time for abduction by using cliff sensors to assess if abduction has happened, thus increasing real-time localization performance. When searching the nearest neighbor obstacle distance, the optimized nearest neighbor search algorithm will sort by the distance from the lookup point to the node's hyperplane, then continue to search the node with the highest priority, and when the set search time is exceeded, directly return the current nearest neighbor distance as the approximate nearest neighbor. Overall, the algorithm has significant improvement in the pre-processing localization phase and good accuracy improvement and time reduction in the real-time localization phase.