图 书 馆
7.运动的合成和控制学习（联合国）打结变形线性对象7. Motion Synthesis and Control Learning for (Un)Knotting Deformable Linear Objects |
|
| 课程网址: | http://videolectures.net/ijcai09_prabhakaran_mscludlo/ |
| 主讲教师: | Sandhya Prabhakaran |
| 开课单位: | 巴塞尔大学 |
| 开课时间: | 信息不详。欢迎您在右侧留言补充。 |
| 课程语种: | 英语 |
| 中文简介: |  |
| 课程简介: | This research deals with motion planning and control for Deformable Linear Objects (DLOs). It is still a complex task to get a robot manipulate a rope or cloth. To realise this vision of getting a robot to handle dexterous objects, we have taken the simplest object i.e. a DLO for our study purpose. The operations performed on the DLO are knot-(un)tying. The DLO is thus parameterised as a Knot and we make the DLO (un)tied into various knot types. The mathematical branch of Knot Theory is extensively used here to realise the Knots. The configuration space that the Knot can move is computed by the Knot Energy. We use the Minimum Distance Knot energy here. With this, we create a hierarchical graph structure with nodes corresponding to optimal knot configurations obtained by optimising this Knot energy functional. Thus by navigating this graph, we are able to (un)tie various knots. The study looks into 3 simple and 2 complex knot types. Motion control while (un)tying is also brought about using the SARSA(λ) [Reinforcement Learning] algorithm. The motion planner is resilient to perturbations as well. Thus by devising Knot Energy together with SARSA(λ), we have built a multi-scale, reactive knot (un)tying motion planner. Results show that our method is incredibly faster than normal Probabilistic and feedback control methods. |
| 关 键 词: | 可变形线性对象; 运动规划; 最优结构型优化 |
| 课程来源: | 视频讲座网 |
| 最后编审: | 2021-12-23：liyy |
| 阅读次数: | 51 |