番茄粘弹性参数机器人抓取在线估计  

Online Estimation of Tomato Viscoelastic Parameters during Robot Grasping

作　　者：周俊[1] 张娜[1] 孟一猛 王明军[1] 

机构地区：[1]南京农业大学江苏省智能化农业装备重点实验室,南京210031

出　　处：《农业机械学报》2017年第8期26-32,共7页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(31471419);高等学校博士学科点专项科研基金博导类项目(20130097110043);浙江省自然科学基金项目(LY17F030006)

摘　　要：为了使采摘机器人在抓取过程中能够对被抓果蔬的粘弹性力学参数进行快速估计,实时优化抓取过程,减少末端执行器对被抓取对象造成机械损伤,以抓取力、变形量、作用时间为输入,建立了番茄粘弹性参数估计的人工神经网络模型。运用质构仪蠕变试验所测的力、变形和时间,以及粘弹性参数E_1、E_2、η_1、η_2作为训练数据集,确定了人工神经网络的拓扑结构和参数,并测试了网络模型的粘弹性参数估计性能。利用二指机器人末端执行器对随机番茄样本进行抓取试验,并在抓取过程中用此模型来在线估计粘弹性参数。通过与质构仪的实测值进行对比发现,当时间t≥0.2 s时,各参数的估计值与实测值之间的相对误差均在25%以内,并根据0.2 s时得到的粘弹性参数对机器人抓取力范围进行了估计。结果表明,利用此方法在机器人抓取过程中可以对被抓番茄粘弹性特性进行估计,为在线优化抓取力提供了依据。When a picking robot is able to quickly estimate the viscoelastic parameters of the fruits and vegetables in the process of grasping,an optimization of the grasping process in real time can be carried out and the mechanical damage caused by the end-effector can be alleviated. Artificial neural network（ ANN） model of tomato viscoelastic parameters estimation was established by using grasping force,deformation and acting time as inputs. The force,deformation and time measured by creep test with texture analyzer,as well as the viscoelastic parameters（ E_1,E_2,η_1,η_2） were used as the training data set to determine the topological structure and parameters of the artificial neural network. Then performance of the network model was tested. A two finger robot end-effector was applied to grasp tomato samples selected randomly,and the ANN model was used to estimate the viscoelastic parameters online during the process of grasping. Compared with the measured value by texture analyzer,when time was more than or equal to 0. 2 s,the relative error between the estimated value and the measured value were less than 25%,and according to the viscoelastic parameters obtained from the 0. 2 s time,the range of the robot＇s grasping force was estimated. The results showed that the method could be used to estimate the viscoelastic properties of the grasped tomatoes during the robot grasping process,which provided the basis for the online optimization of grasping force.

关 键 词：机器人抓取 番茄 粘弹性参数 蠕变试验 人工神经网络 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]