Control Technology Of Multi Sensor-based Semi-autonomous Telerobot

With the implement of space exploration plan,the telerobot technology was developed in the last century.In the recent years,the application of telerobot has been gradually applied in the civil fields,and it was introduced into varies advanced manufacturing industry for the design,manufacture,exhibition of product,which improves the productivity and product consistency obviously.Moreover,the development of novel sensor technology and AI technology make the telerobot more autonomous.In this thesis,a novel multi sensor-based semi-autonomous telerobot is introduced,which can be used in the civil field,replacing human to execute some dangerous tasks such as polishing,classifying and assembling etc.This system is composed with local module,communication module and remote module.The local module includes Human-Computer Interface(HCI)facilities and local computer,while the remote module includes robot,cameras,6-axis force sensor and remote computer etc.Compared with the purely manual system,this system has better performances in the aspects of accuracy and efficiency.Compared with the purely autonomous system,this system can execute task more flexibly during the industrial production.The research contents in this system include: 1)Two cameras are used in this system.The deep image of the work platform is acquired by binocular camera,which can be used to render the virtual reality environment.Another image is also acquired by monocular camera,which is used for target matching,classifying and polishing.This system complete both the HCI task and target matching task respectively according to the advantages of different cameras.2)Considering the telerobot is usually used to execute some complex tasks,this system works in two mode: Teaching Mode and Auto Mode.In the Teaching Mode,operator controls the telerobot to execute task,then the system creates the dynamic template for unfamiliar target and record the teaching data.In the Auto Mode,the robot execute task,such as target classifying and polish,based on the dynamic template and teaching data,which help robot adjust the polishing force,grasping force etc.3)A SB-GrabCut algorithm is proposed to segment target from texture-rich background for template creation.Compared with the traditional GrabCut algorithm,SB-GrabCut can segment a more accurate template with less iteration times and less time consuming.4)A MB-NCC algorithm is proposed to match the target in texture-rich image.Compared with the traditional PP-NCC algorithm,MB-NCC is more robust against the noise.Especially for the matching in texture-rich environment,MB-NCC has more advantages,because it can create a binary map for each template to eliminate interfere from background pixels.5)Considering that the same kind of casting often has different texture,an edge featurebased matching algorithm,named GB-NCC,is proposed.As the MB-NCC narrated above,this algorithm also use the binary map and polar transformation to eliminate interfere from background pixels.Compared with the traditional edge feature-based matching algorithm,the time consuming of GB-NCC is less.