| 摘要: | 機器手臂前端的夾爪一直是技術研發重點之一，因為傳統機器人夾爪的夾取力道是固定的，一 旦被夾取物為隨機樣式，有可能大小不同、形狀不同、重量不同、硬度不同等等時候，可能造成被 夾取物變形甚至損毀，因此仿生夾爪的設計就很受到矚目。另方面，類神經網路與基因演算法目前 已被成功地應用在各個不同的領域上，基本上兩者均有增加系統智慧，透過訓練，可學習輸入與輸 出對應關係的能力；而基因演算法的最大優點是當它在尋找最佳解的過程中，它可以藉由基因的交 配與突變動作，來避免陷入局部性的最佳點，而找到全域最佳解的機會。 為延續並改善已完成的專題研究計畫(NSC102-2221-E-034-005-往復式足型機器人導航系統與仿 生夾爪之開發)機器夾爪的部分，此計畫擬發展一套以類神經網路與基因演算法為基礎的仿生機器夾 爪，作為寵物機器人之用。計畫中將增加壓力與彎曲感測器數量，以提高其靈敏度，並以雙層循環 式類神經網路建模與基因演算最佳化的方法，控制仿生夾爪的動作，計畫內容包括：（1)仿生機器夾 爪機構設計與製作；（2)仿生機器夾爪之類神經網路控制器設計與製作；（3)仿生機器夾爪之混合式類 神經網路控制器設計與製作；（4)在寵物機器人上加裝仿生機器夾爪：由於寵物機器人具有陪伴高齡 長者的功能，但目前市面上的產品，大多沒有機器夾爪，因此擬將上述成果，應用於寵物機器人上。 為完成上述構想，擬提出一年期的進階計畫，在已有的基礎上，進行新型仿生機器夾爪機構設 計與製作以及智慧型控制器設計、程式撰寫與整合測試。此研究計畫，整合微處理機、控制、仿生 材質、感測器、機構設計與製造、寵物機器人，可培訓一些具有實務能力的學生，也希望能帶給國 內有意開發服務或娛樂型型機器人的業者一些助益，預計完成的工作包括： (1)雙層循環式與放射狀類神經網路及基因演算法則推導。 (2)撰寫Matlab程式進行測試與效能評估。 (3)仿生機器夾爪機構推導、設計與製作。 (4)感測器選擇與安裝。 (5)仿生機器夾爪控制器硬體設計與實現。 (6)撰寫仿生機器夾爪驅動程式，測試與除錯。 (7)寵物機器人之應用。 (8)整合測試與除錯。
The front end of the robot arm has been one focus of technology development. Once the gripping objects are random patterns (different sizes, different shapes, different weights, different hardness and so on), the distortion or damage may be caused due to the fixed traditional jaw gripping force. On the other hand, neural network and genetic algorithm have currently being successfully applied in various fields, both of them can substantially increase the intelligence of the system and can learn the corresponding relation between the input and output data through the training process. The biggest advantage of genetic algorithm is that it can find the global optimal solution and avoid falling into local optimal point via the crossover and mutation operations. To continue the research of the bionic handling gripper in the previous project (NSC102-2221-E-034-005- Development of navigation system and bionic handling gripper for reciprocating-foot robot), a new neural network and genetic algorithm based bionic handling gripper will be developed in this project. In the proposed gripper, more pressure and bending sensors will be utilized to increase its sensitivity and the recurrent neural network as well as the genetic optimized method will be adopted in the controller to handle its action. The project contents include:⑴ The mechanism design and manufacture of the proposed bionic handling gripper. (2) Design and implementation of the double-layer recurrent neural network based controller. (3) Design and implementation of hybrid intelligent controller. (4) The application of the proposed gripper to the smart pet robot. To implement this idea, an advanced project is proposed to continue to research the new mechanical structure, control method and application of the bionic handling gripper. This project is divided into two major parts and the execution duration is one year. The first part is the derivation, design and manufacture of the gripper mechanism. The second part is the derivation, design and implementation of intelligent controller. The technology of this project is the integration of microprocessor, control, material, sensor, mechanism and smart pet robot. This project can give the students some practical training and the robot industry some help. Following works will be completed in this project: ⑴ The derivation, design and implementation of the control laws which are based on the double-layer recurrent, radial basis neural network and genetic optimized algorithm. (2) Write the Matlab program to evaluate the performances. (3) The derivation, design and manufacture of the proposed bionic handling gripper mechanism. (4) The selection and installation of the pressure and bending sensors. (5) The design and implementation of the hardware of the controller. (6) Write, test and debug the drive program. (7) Application to the smart pet robot. (8) Integrated testing and debugging. |
