Study On AGV Autonomous Positioning And Path Planning Strategy Based On Visual SLAM

With the continuous breakthrough of technology in the field of artificial intelligence,the development of autonomous mobile robot technology is getting faster and faster.As an important branch of robots,the development of the Automated Guided Vehicle(AGV)industry has ushered in a huge breakthrough.For the traditional Global Positioning System(GPS),when it is applied to the positioning function of indoor robots,the positioning accuracy of the robot in the indoor limited area has certain uncertainty.At the same time,when implementing the path planning and guidance functions,traditional methods such as preset orbits and tracking QR codes have a high dependence on external auxiliary facilities and poor flexibility.When the placement of items in the space changes dynamically,the effective utilization of space is low.In view of the above problems of AGV positioning and path planning,this paper combines the mainstream algorithm framework ORB-SLAM of simultaneous localization and mapping(SLAM)to implement the positioning function of AGV,and studies the intelligence based on reinforcement learning.Algorithmic path planning strategy.The research work is as follows:First,this article studies the working principle of the visual SLAM system.Since the RGB-D sensor can obtain both color information and depth information,the RGB-D camera Kinectl produced by Microsoft is employed as the only external sensor to collect environmental information data.This paper analyzes the working principle and imaging model of Kinectl,and introduces the implementation principle of front-end odometer based on vision SLAM in detail.Aiming at feature point extraction and matching,combined with ORB algorithm to improve the traditional feature description defects.The camera motion is roughly estimated based on the information between the two frames,and the pose is solved by using Random Sample Consensus(RANSAC)and Iterative Closest Points(ICP)algorithms.Secondly,the AGV positioning problem is studied.The classification of positioning problems is discussed,and corresponding analysis is performed for different positioning problems.Several commonly used positioning algorithms are discussed in detail,including Markov positioning algorithms based on probability filtering,Monte Carlo positioning algorithms based on particle filtering,and adaptive Monte Carlo positioning algorithms.The grid map is used as the environment modeling map,and the algorithm implementation of the grid map is discussed in detail.Finally,this paper applies reinforcement learning to robot path planning tasks and implements path planning algorithms in static and dynamic environments,respectively.For static environment,this paper adopts grid method to model the space,and uses Q-Learning algorithm to train the robot to separate strategy,and simulates the robot path planning in grid environment.For dynamic environment,based on the traditional artificial potential field method,this paper proposes a reinforcement learning path planning algorithm based on potential field.By rationally designing continuous state and action space and reward function,a Markov decision process is established.The DDPG algorithm optimizes training and implements robot path planning.The comparative analysis of simulation experiments proves the effectiveness and stability of the algorithm.