Vision-based initial localization of AGV and path planning with PO-JPS algorithm

In recent years, robot path planning has gained high attention. The traditional adaptive Monte Carlo localization (AMCL) has such problems as limitations in global localization, and incomplete path and time-consuming problem in path planning due to too much calculation of meaningless nodes by the jump point search (JPS) algorithm. In view of the above problems, this paper proposed a method for vision-based initial localization of automated guided vehicle (AGV) and path planning with (pruning optimization) PO-JPS algorithm. The core contents include: vision-based AMCL localization module and improved JPS algorithm based on pruning optimization. Firstly, Oriented FAST and Rotated BRIEF (ORB) features are extracted from the images collected by vision, and coordinates are localized with the features, coupled with the initial map by laser SLAM, to construct a bag-of-words (BoW) library of features. The key frame most similar to the current one is obtained by comparing the similarity between the current and historical frames in the BoW library. The Euler transformation between these two frames is calculated, to carry out pose estimation. This pose, as an initial value, is provided to the AMCL for particle iteration. Secondly, in the path planning stage, an improved JPS algorithm based on pruning optimization is proposed, and a strategy that the repeated intermediate inflection points in the complemented path after pathfinding are deleted is designed. Therefore, while a complete path is obtained, the calculation workload and memory consumption for meaningless nodes during node extension are reduced successfully, and the efficiency of the pathfinding algorithm is raised. Finally, verification of the method proposed in this paper is completed through a large number of simulations and physical experiments, which saved 17.7% of the time compared to the original JPS algorithm and 279.6% to the A* algorithm.