论文部分内容阅读
An autonomous exploration and navigation system of a mobile robot in an unknown environment is a fundamental challenge in the field of a mobile robot.Many researchers have been researching autonomous exploration and navigation systems of a mobile robot for personal,service,military purposes.This system is generally used for too dangerous tasks for humans or required high precision and repeatability.Particularly,mobile robots are used to exploring unknown environments such as space,underwater,and disaster area.In order to perform this,mainly the Simultaneous Localization and Mapping（SLAM）system and motion planning algorithm are used.Also,A grid-based search or artificial potential fields,sampling-based algorithm,image processing algorithm,control logic,and computer vision are designed and integrated into the system.This thesis proposed an Autonomous Exploration and Navigation System based on a Semantic SLAM application for a mobile robot.It combined techniques such as image processing algorithms,motion planning method,and fuse of multi-data,Semantic SLAM,control logic,obstacle avoidance,search-exploration point.The first approach is to implement the SLAM System using the Semantic SLAM application,which is based on ORB-SLAM2 with YOLOv3,to get rid of dynamic objects in the scene while Mapping and localization on a mobile robot.Second,asynchronous data fusion in Extended Kalman Filter（EKF）formed by the wheel odometer and visual odometer is implemented for the optimized localization and the Navigation system.Third,in order to implement an exploration in an unknown environment,an exploration point for an unknown region is obtained by combining 2D occupied mapping information,which is a downprojected map data from Map Points information using Point Cloud,with image processing,which is the Canny Edge Detection algorithm and the bitwise algorithm.It is called "Frontier Point Detection" in this thesis.Fourth,keep searching newly obtained frontier points through the Frontier Detection algorithm in real-time,and update them to the Motion Planning system.It is called "Fast Frontier Point Detection" in this thesis.Fifth,Motion Planning for Exploration and Navigation are implemented by PRM,RRT,RRTstar algorithm using Map Points and given target pose obtained by Frontier-point detection or user,and a mobile robot’s characteristics.Six,designing the Obstacle Avoidance algorithm based on 2D occupied mapping information.Lastly,the autonomous exploration and navigation system for a mobile robot is implemented by integrating with all the above-presented methods,Semantic SLAM application and Fast Frontier-based Detection,Motion Planning,and Obstacle avoidance,Data Fusion.The mobile manipulator with Mecanum wheel chassis and a universal robot manipulator with a Kinect v2 camera mounted on top of the Manipulator is used to evaluate this approach.Furthermore,using Turtblebot,which has an RGB-D camera in the Gazebo Simulation environment to validate our system.A computer with intel i9 CPU performs the proposed method,Nvidia GTX Titan 2080 ti,and 32 GB,and validated by Simulation environment as Gazebo and TUM2 and ROS kinetic in ubuntu 16.04 LTE.