Development Of Two-wheel Differential Visual Localization AGV Trolley

Automated Guided Vehicle(AGV),as the core of intelligent logistics in smart manufacturing,has become a hot topic of research nowadays.This paper developed a two-wheel differential speed AGV based on vision localization to address the problems of low accuracy,poor robustness,harsh environmental requirements and poor flexibility of the traditional AGV guidance method,which cannot meet the demand for high accuracy,strong robustness,high intelligence and high flexibility of AGVs from intelligent manufacturing.The AGV uses Visual Simultaneous Localization and Mapping(VSLAM)algorithm to localize itself in real time,and a fuzzy control algorithm to control the rotation speed of the left and right wheels to form a differential speed,thus making itself travel along a preset path.Visual SLAM can locate only based on features in the environment,freeing it from the need to artificially set guide belts or wayfinding objects.AGVs using this method of locating can operate in unknown environments and work in a wider range and with higher ubiquity.The paper investigates the two-wheel differential speed AGV with visual localization from both theoretical principles and implementation aspects,and completes the design and construction of this AGV.The main work of this paper are as follows:(1)The current development status of key technologies for AGVs at home and abroad is introduced,and the advantages and disadvantages of the main guidance technologies for AGVs are analyzed.The basics involved in AGV and visual positioning are introduced in detail,including the drive mode and kinematic model of AGV,the control algorithm of AGV,the model of pinhole camera and binocular camera,the localization solution and optimization algorithm,etc.(2)Motion controllers of AGV in this paper are designed according to two control algorithms commonly used for AGV.And Simulink of Matlab is used to simulate and model the path-following of AGV.By comparing and analyzing the path-following simulation results of the two designed AGV motion controllers,the performance of the two controllers is verified,and the fuzzy controller with better performance is selected as the controller of AGV.(3)An image dynamic region detection algorithm is proposed.The algorithm first quickly detects a small number of feature points and uses multi-view geometry to check their motion consistency and detect dynamic feature points.The images are segmented using the improved superpixel algorithm in this paper,and the superpixel boundaries are classified by disparity.Finally,using dynamic feature points and superpixel boundary types,the region where the dynamic object is located can be obtained.Experiments show that this dynamic region detection algorithm achieves better results in both dataset and real dynamic environment.(4)For the SLAM problem in dynamic environment,ORB-SLAM2 is improved using the dynamic region detection algorithm proposed in this paper,and the improved SLAM algorithm is compared with the original ORB-SLAM2 and Dyna SLAM for experiments.The experimental results realistically show that the improved SLAM algorithm in this paper has similar accuracy to Dyna SLAM in dynamic environment,but higher than the original ORB-SLAM2.(5)A two-wheel differential speed AGV based on visual localization was built,and the path tracking performance and the feasibility of various algorithms of the AGV were verified through path tracking experiments in both static and dynamic environments.