BIM Based Localization And Environment Sensing System For Autonomous Interior Finishing Robot

Interior finishing is a kind of manual labor with high repetition and intensity.With the increasing cost of human resources,how to reduce or even replace human resources has gradually become a practical problem facing the decoration industry.In recent years,with the rapid development of artificial intelligence and robot technology,many traditional industries have realized automation and intelligence through the application of these cutting-edge technologies.The application of robot technology to the decoration industry has become the only way to solve the problem of human cost and promote the development of the industry.However,the application of robot technology in the field of interior finishing is still very limited.One of the important reasons is that the robot's ability to locate itself in the construction scene and to perceive the construction scene is weak.The decoration scene is different from the general indoor robot application scene,and its most significant features are: 1.The environment is not completely unknown,there are prior information such as construction drawings;2.The construction scene will change with the construction progress;3.The accuracy requirements for environmental perception are high.The current environmental awareness technology is not fully applicable to the application scene of interior finishing robot,so this paper develops a set of interior finishing robot positioning and environmental sensing system based on BIM(Building Information Model).The system combines BIM,the most advanced information technology in the construction industry,and SLAM,the most advanced localization and mapping technology in the robot field,and explores the new direction of the combination of robot technology and traditional decoration industry.The system obtains the structural information of construction environment through BIM files.Because BIM features,the prior information covers all stages of construction,this paper cleverly uses redundant information and Monte Carlo positioning algorithm to realize the location of variable structure scene.After the robot position information is obtained,the visual features of the wall are obtained by using the visual feature enhancement technology and ZED-Mini stereo camera,and the robot position information obtained by BIM is fused to complete the three-dimensional information perception of the interior finishing environment and obtain the three-dimensional visual feature map of the interior finishing environment.In order to realize the specific process,the robot needs to perceive the semantic information of the environment,such as the wall defects.Therefore,this paper usesconvolutional neural network to identify the wall defects,and combines with the slam framework to calculate the defect location and mark it in the 3D visual feature map for subsequent use.In this paper,we also explore the dense semantic perception of the interior finishing robot,complete the pixel level semantic segmentation of the environment through the deep neural network,and combine the results with slam to build a three-dimensional occupation grid map that can be used for robot behavior planning,which lays the foundation for more complex and larger-scale independent construction of the interior finishing robot.