当前位置: X-MOL 学术Mechatronics › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Template-based imitation learning for manipulating symmetric objects
Mechatronics ( IF 3.1 ) Pub Date : 2021-07-10 , DOI: 10.1016/j.mechatronics.2021.102609
Cheng Ding 1 , Wei Du 1 , Jianhua Wu 1 , Zhenhua Xiong 1
Affiliation  

With the rise of small batch production, the need to shorten the development cycle has become urgent. This work aims to allow people without expert knowledge to program the robot in a single demonstration to reduce development costs. The idea of this work is to record frame relations between objects during the demonstration as templates. Then the robot is programmed by the template sequence to simplify the programming process and shorten the development time. Since symmetry will cause uncertainty in pose estimation during execution, a symmetry detection module is developed to detect the rotational symmetry of the objects through inertia tensor and all the rotation matrices are acquired using group theory. After that, a template matching module is designed to measure the distance between the current pose and possible target poses to choose the proper target pose which matches the template to avoid unreachable robot poses and accelerate the executing process. Two pick-and-place tasks, a stacking test, and a sorting task are carried out to verify our system and the results show that our method is feasible and efficient.



中文翻译:

用于操作对称对象的基于模板的模仿学习

随着小批量生产的兴起,缩短开发周期的需求变得迫切。这项工作旨在让没有专业知识的人在一次演示中对机器人进行编程,以降低开发成本。这项工作的想法是将演示过程中对象之间的框架关系记录为模板。然后通过模板序列对机器人进行编程,以简化编程过程,缩短开发时间。由于对称性会在执行过程中导致姿态估计的不确定性,因此开发了对称性检测模块,通过惯性张量检测物体的旋转对称性,并使用群论获取所有旋转矩阵。之后,模板匹配模块用于测量当前位姿与可能的目标位姿之间的距离,以选择与模板匹配的合适目标位姿,以避免无法达到的机器人位姿并加速执行过程。进行了两个拾放任务、堆叠测试和排序任务来验证我们的系统,结果表明我们的方法可行且有效。

更新日期:2021-07-12
down
wechat
bug