Font Size: a A A

Cobot's Manipulation And Implementation Research Based On The Sense Of Vision And Force

Posted on:2022-09-03Degree:DoctorType:Dissertation
Country:ChinaCandidate:S ZhangFull Text:PDF
GTID:1488306572475154Subject:Mechanical engineering
Abstract/Summary:PDF Full Text Request
The current mathematical modeling for the robots' planning is very challenging.And the robustness of the available mathematical models in dynamic task is low;so does the efficiency of maintenance.Thus,this paper developed the manipulating cell combining the intelligence of cobot with that of human.And the modules of this intelligent manipulating cell are designed to provide a set of systematic and effective method which can improve the efficiency of task planning and new task learning.Also,this method can enhance the robustness in the dynamic task of cobot.All in all,this study is of great significance for improving the performance of manipulation of cobots.To realize the grasping force control of the deformable object grasping for a sensor-less cobot,this paper introduced the fuzzy hardness(FH)for imprecise grasp force evaluation.In addition,a method to infer the FH of objects was proposed based on supervised learning via long short term memory network(LSTM).In this method,the deformation of objects related to the close degree of gripper was treated as a key variable and measured via visual methods.Based on the measured deformation data,LSTM was introduced to conduct supervised learning synchronously.Then several predicted deformation curves can be obtained through these LSTM blocks.Subsequently,the FH of objects would be clear when the errors between measured data and the predicted ones was calculated from the curves.The FH can be used to adjust the close degree of gripper.That can decrease the deform degree of object.The verification experiments showed that with the method,the FH can be inferred quickly in the process of grasping.Then the fuzzy grasping force control can be realized for force sensor-less cobots based on the visual information.There are more and more task-level commands for cobots,which introduces plenty of programming work.Hence a task planning algorithm based on task semantic object matrix(SOM)was introduced in this paper.With this strategy,petri net was applied to decompose the cobots' complex operation tasks into sequential subtasks,which can be practical for finite state machine.And the SOM related to sequential subtasks was generated synchronously.Based on the SOM,the point cloud segmentation,collision geometry modeling and the transformation between targets and obstacles can be conducted for cobots' operations.The experimental results indicate that the SOM algorithm can improve the understanding ability of tasks and the robustness of operation in dynamic scenarios for cobots.The efficiency of the planning and the selecting of grasping points via mathematical model is low.It cannot satisfy the trend of the personalized customization and small-lot production in manufacturing industry.Therefore,the imitating CNN method was developed in this paper to deal with the problem aforementioned.In this method,the postures of objects were associated with the grasping postures of the human via calculating geometrical intersection.Then the imitating CNN net was designed to make cobots learn grasping experience from visual information.Also,the experimental results indicated that this algorithm can help cobots learn to grasp one object within 20 minutes if the grasping points data of human are collected.The safety planning algorithm based on the sense of vision and force was proposed in this paper.It can realize the collaborative operations more efficiently at the premise of ensuring the safety of humans initiatively.In this algorithm,the digital dynamics model of human was built through visual information.And the minimum distance between human and cobot was calculated in real-time.Subsequently,the minimum distance was converted into virtual force which can establish the link between cobot and human.Besides,the value and the time duration of virtual force were employed to evaluate the safety status of human-robot system.Finally,cobot can conduct safety planning initiatively according to the safety status.In this way,humanrobot cooperation like contact maintenance can conduct in a non-stop mode.That can improve the efficiency of production line.Based on the above research results,a prototype system of this paper developed the manipulating cell which combined the intelligence of cobot with that of human.The related functional modules of the prototype system were verified through two experiments.One was the maintenance experiment of spray which involved physical contact.The other is the imitating grasping experiments of shoe sole.In the verification experiments of cobot's maintenance involving physical contact,the average response time from the detection of the distance change of human-robot to the adjustment of the cobot is 0.05 s.Moreover,this algorithm can maximally increase efficiency by 26.4% compared with the traditional stop-maintenance-restart mode.Experimental results show that the grasping planning efficiency of the imitating CNN method is improved by 26.1% compared with the pick-and-place module of ROS.
Keywords/Search Tags:cobots, vision and force, task semantic object matrix, fuzzy hardness inference, imitating learning, human-robot safety
PDF Full Text Request
Related items