Research On Key Technology For Chemical Robot Based On Colored Solution

Robot assisting chemical experiment is an important research of robot technolog y.The widely used robot recognition and grasping algorithms are generally aimed at o bjects with low surface smoothness or opacity.Because of the particularity of transpar ent solution,it is difficult for vision sensors to accurately obtain its deep information,which leads to the inability to correctly evaluate the position and posture,and complet e the recognition and grasping of the grasped transparent object.Due to the large num ber of transparent container operations are needed in chemical experiments,it is urgen t to study the key technologies of chemical robot grasping colored solutions.Aiming at the shortcomings of high-precision methods for obtaining deep inform ation of colored solution containers in chemical experiments,deep information repair theory is introduced,and an object recognition and grasping method based on deep inf ormation repair is established focusing on the application requirements of color soluti on containers recognition and grasping in chemical experiments,which provides theor etical basis and method support for obtaining deep information of objects with transpa rent objects,and effectively improves the operation accuracy and efficiency of chemic al robots in color solution experiments.The main contents of this paper are as follows:(1)A method of repairing the deep information of transparent objects in colored solution is proposed.After obtaining RGB-D images of transparent objects with color ed solution by camera,the geometric information such as surface normal and occlusio n boundary of transparent objects is predicted based on RGB images,and the actual d eep pixel value of target objects is predicted by Conv Ne Xt network based on the abov e information,so as to achieve the restoration effect of deep information of transparen t objects with colored solution.Compared with other deep information restoration met hods on Cleargrasp data set,it is proved that the new deep information restoration mo del is concise and effective,and has good generalization.(2)Based on the restoration deep information,the model of clutching transparent body in colored solution is established.The better grabbing model network framewor k is chosen,and three different input grabbing network models under the network fra mework are proposed.The grabbing model based on deep information,fusion RGB-D and multi-channel RGB-D.The three models are trained on Cornell data set respectiv ely,and verified on the extended cleargrasp data set,and the best grabbing model is o btained by comprehensive analysis.(3)Based on Baxter robot,the experimental verification platform of colored solu tion recognition and grasping algorithm is built.Based on the recognition algorithm of colored solution,an experimental platform is built by experiments.The results show t hat the new algorithm is stable and effective in recognition and grasping of colored so lution.The comparison experiments of other recognition and grasping methods are co mpleted on the same experimental platform,which further verifies the effectiveness of this method.