智慧型服務與人形機器人

(1)

1

國立成功大學「邁向頂尖大學計畫」

　　延攬優秀人才工作報告表

NCKU’s “Aim for the Top University Project”

Work Report Form for Distinguished Scholars

□續聘continuation of employment █離職resignation

105 年 8 月 29 日更新受聘者姓名

Name of the Employee 林國榮 ▓男女

Male Female

聘期 Period of Employment

From 2016 年(y) 01 月(m) 01 日(d) To 2016 年(y) 08 月(m) 31 日(d) 研究或教學或科技研發與

管理計畫名稱 The project title of research,

teaching, technology development and management

智慧型服務與人形機器人

計畫主持人

（申請單位主管）

Project Investigator (Head of Department/Center)

李祖聖教授

補助延聘編號

Grant Number HUA 105-3-4-077

一、研究、教學、科技研發與管理工作全程經過概述。（由受聘人填寫）

Please summarize the entire research, teaching, or science and technology R&D and management work process (To be completed by the employee)

這段聘任期間主要是在先進智慧型機器人與系統實驗室從事頂尖計畫之執行。計畫之題目是 “智慧型服務與人形機器人＂。在服務方面協助老師整理及設計針對與台達研究院合作之產學合作案 [此計畫題目是“未知系統快速建模與控制器參數自動調校: 以聚合脢連鎖反應儀為

應用平台”]，而我是負責系統建模這部分。應用递推增廣最小二乘法可以將所量測出之輸出及輸

入的數據系統判別出來以得到一階時延系統的差分方程式。再將此轉換關係的差分方程式應用於模糊控制器的相關設計應用。

另外在此計畫中主要是針對機器手臂(robotic manipulators)之應用提出新的相關理論與演算方法來作為控制器之設計。主要方法著重在應用深度學習(deep learning)或相關學習法則來作學習之研究、應用類神經網路於機器手臂之控制應用、應用模糊控制於機器手臂之控制應用及整合類神經網路與模糊控制及學習法則於機器手臂之控制應用。根據目前所刊登於頂尖期刊的研究資料顯示，目前比較新型之控制應用如適應類神經網路控制，模糊滑模控制器設計，適應模糊控制及反演(Backsteeping)控制等應用於機器手臂的控制。由於機器手臂不只是一種高度非線性，高度偶和及時間改變系統，而且也存在許多結構與非結構之時間不確定性。傳統的機器手臂其數學動態方程式表示如下:

M(q)qC(q,q)qG(q) d

其中 q ， q 及 q分別代表手臂連桿的位置，速度及加速度。M(q)是代表慣性矩陣，C( qq, 是代) 表柯里斯力及離心力，G q 是代表與地心引有關之力矩相量，( )  是代表馬達力矩之輸入向量及d 是代表外部擾動信號。由目前文獻可以看出慣性矩陣M(q)，柯里斯力及離心力C( qq, 及與地心) 引有關之力矩相量G q 從實際的應用觀點來說是不完全知道的，因此必須應用適應控制定律來( )

(2)

2

建立以上的估測向量矩陣Mˆ q( )，Cˆ(q,q 及) Gˆ q 。經由此方向思考及設計可以推導出一些控制機( )

器手臂的方法，所提出的方法，與頂尖期刊文獻^*[A]相比可以得到較佳的結果。目前將整理出二

篇期刊發表。

＊文獻 [A] W. He, Y. Chen and Z. Yin, “Adaptive neural network control of an uncertain robot with full state constraints”, IEEE Trans. Cyber., vol. 46, pp. 620-629, 2016.

談到研究成果部分，目前已整理好兩篇論文準備修改並且投稿。一篇論文題目是“Dynamic learning for robust adaptive tracking control of robotic manipulators”此篇論文之英文摘要大致說明如下：

In this paper, we propose dynamic learning for robust adaptive tracking control with a guaranteed H performance for robotic manipulators with external disturbances. A radial basis function (RBF) neural network (NN) system is addressed to learn a desired dynamic via gradient descent training rule.

The learned knowledge can be stored in memory. We can reuse the learned knowledge to approach the desire dynamic. Based on the Lyapunov stability theorem, the proposed motor torque via robust adaptive tracking control design can achieve the desire tracking control performance. Furthermore, the developed H^ control performance guarantees that the influence of external disturbance is as small as possible. The model parameters are assumed not known. Two adaptive control design methods are addressed for the estimated model parameters. A two degree of freedom robotic system is given to illustrate the validity of the proposed scheme.

主要結果: 所得到之追蹤誤差是明顯最小。

另一篇論文題目是“Neural network based adaptive sliding mode control for robust tracking control of robotic manipulators”此篇論文之英文摘要大致說明如下：

In this paper, we propose neural network based adaptive sliding mode control for robust tracking

(3)

3

control with a guaranteed H^ performance for robotic manipulators with external disturbances.

Radial basis function (RBF) neural network (NN) is adopted to approximate a desired dynamic, unknown parameters, and disturbances. Based on the Lyapunov stability theorem, the proposed adaptive sliding mode control via robust racking control design can achieve the desire tracking control performance. Furthermore, the developed H^ control performance guarantees that the influence of external disturbances is as small as possible. Model based adaptive sliding mode control (ASMC) design and NN-based ASMC design are addressed for robotic manipulators with external disturbances.

A two degrees of freedom robotic manipulator is shown to illustrate the validity of the proposed scheme.

主要結果: 所得到之追蹤誤差是明顯最小。

※目前有具體的研究成果整理成投稿文章如下:

[1] T.-H. S. Li, and K.-J. Lin, “Dynamic learning for robust adaptive tracking control of robotic manipulators,” IEEE Trans.(Submitted).

[2] T.-H. S. Li, and K.-J. Lin, “Neural network based adaptive sliding mode control for robust tracking control of robotic manipulators,” IEEE Trans.(Submitted).

二、研究或教學或科技研發與管理成效評估（由計畫主持人或單位主管填寫）

Please evaluate the performance of research, teaching or science and technology R&D and management Work: (To be completed by Project Investigator or Head of Department/Center)

(1)是否達到延攬預期目標？

Has the expected goal of recruitment been achieved?

聘任林博士主要是協助頂尖計畫: “智慧型服務與人形機器人＂的執行，也期望有國際期刊論文的產出，因此也有達到延攬之預期目標。

(2)研究或教學或科技研發與管理的方法、專業知識及進度如何？

What are the methods, professional knowledge, and progress of the research, teaching, or R&D and management work?

林博士在此段聘任期間研讀相關頂尖的期刊，將所學習的方法應用在研究上，也提出不同的

(4)

4

方法來解決機器手臂之控制問題。因此研發方法、專業知識及進度是如預期平順的。

(3)受延攬人之研究或教學或科技研發與管理成果對該計畫(或貴單位)助益如何？

How have the research, teaching, or R&D and management results of the employed person given benefit to the project (or your unit)?

所提出的研究成果將整理成論文發表，因此對該計畫及系上的研究產出有很大的助益。

(4)受延攬人於補助期間對貴單位或國內相關學術科技領域助益如何？

How has the employed person, during his or her term of employment, benefited your unit or the relevant domestic academic field?

對該計畫及系上或國內相關學術科技領域助益很大。

(5)具體工作績效或研究或教學或科技研發與管理成果：

Please describe the specific work performance, or the results of research, teaching, or R&D and management work:

整理好的論文將投稿頂尖期刊。

(6)是否續聘受聘人? Will you continue hiring the employed person? □續聘Yes ▇不續聘No 因計劃案已結束及個人生涯規劃，因此離職。

※ 此報告表篇幅以三～四頁為原則。This report form should be limited to 3-4 pages in principle.

※ 此表格可上延攬優秀人才成果報告繳交說明網頁下載。

This report form can be downloaded in http://scholar.lib.ncku.edu.tw/explain/