基於位置與匹配控制之類人形機器人系統

在任何一個環境之中，機器人所在乎的不外乎人與物。就類人形機器人整體架構而言，與人和物接觸最頻繁的就屬其四肢。本篇論文重點為發展PID位置與匹配控制(Proportional-Integral-Derivative Position and Impedance Control)並透過模擬與實驗測試驗證此控制方法兼具精準定位以及安全之效果，以作為類人形機器人四肢底層控制架構。為快速進行PID位置控制之驗證，首先設計一單自由度之機械關節，並以安全為前提設計一無感測器適應剛性致動器SASA (Sensorless Adaptive Impedance Actuator)進行測試。其訴求為希望藉由機構設計與控制器設計來達到機械手臂與人之安全互動。在實驗部分，我們一共進行三部分的實驗，分別為(1)撞擊測試(2) 適應阻抗產生器(3)順應效果。在實驗(1)部分，我們分成兩部份進行測試，靜態壓迫測試以及動態撞擊測試。其中在動態撞擊測試，我們比較了有/無SASA以及無制動器(unactuated)以同樣之速度對dummy進行動態之撞擊的HIC值，結果發現此裝置確實發揮其安全效果。在實驗(2)中，由於位置回授以及積分器的幫助，本控制架構確實能夠讓系統精確的定位，同時適應阻抗產生器能任意設定等速段”速度”與”範圍”以達到安全之訴求。此外，力量觀測器亦消除了輸出扭力的ripple現象，這可讓手臂移動的更加平順。在實驗(3)我們刻意將減速比降低來顯現這個控制架構的特性。然低減速比易造成系統的damping下降，因此我們在原控制架構上加上了微分器來增加阻尼。實驗顯示，單軸PID 位置控制可以提升機器人順應環境之能力。由於匹配控制與PID位置控制在控制架構上是相似的，因此我們也藉由這個實驗了解匹配控制的特性。緊接著我們利用一小型類人形機器人腳部模型來進行PID位置控制與匹配控制器模擬之驗證，此機器人具有32個自由度，傳動方式利用馬達銜接精密之Harmonic Drive做為各軸之驅動源。機器人整體使用板金做為其架構，能夠降低其整體重量。機器人本身配有陀螺儀、壓力感測器，各軸的馬達具有Encoder及Hall Sensor，這些回饋的資訊讓我們可以用來控制機器人之各種動作。我們使用MATLAB進行PID位置與匹配控制模擬，此控制方法結合PID位置控制與匹配控制方法。在腳部與環境無接觸時，使用PID位置控制並利用Genetic Algorithm(GA) and Nelder Mead Simplex Metho(NM)進行各軸較佳的PID參數搜尋達到好的位置控制。在發生接觸後，控制系統切換成匹配控制以便限制垂直方向與腳踝之剛性與施力，同時保持切向有較佳之定位精度與剛性。在初期我們進行兩種步態規畫:等速度及等加速度，其中等加速度符合實現的條件，我們運用原型機進行測試確實能達到我們需要的步態且我們可以任意的決定步長與跨步時間。另外我們藉由線性重力補償器來獲得機器人的靜態穩定步行。此方法在具有物理意義的model上也獲得穩定步行之結果。為展現控制器效果，我們分別將PID位置控制與匹配控制結合在虛擬的機器人上。模擬結果顯示，類人形機器人行走藉由此控制方法能夠較精準的控制卡式座標剛性也因此不會引起不均勻的接觸力量並能夠降低單腳支撐相因為位置誤差所產生與環境的抵抗力量，其效果相較於僅使用PID位置控制是好的。此外，擺動腳使用PID位置控制，其PID控制器參數是由GA與NM所獲得因此能有較佳的系統效能讓其能精確追隨規劃之軌跡。此外，由於是基於力量控制模式並配合反積分終結之機制，因此在與地面接觸時也不會產生過大的交互作用力。未來類人型機器人將以PID位置與匹配控制作為底層之控制架構並進一步發展動態之穩定步行及手臂與環境之安全之互動。

關鍵字

無感測器適應剛性致動器；適應剛性產生器； HIC(Head Injury Criterion) ； PID位置與匹配控制；基因演算法； Nelder Mead單純形法

並列摘要

Since the impedance based control law is similar to the control method that a human uses for motion and locomotion, this thesis attempts to study PID position and impedance control for physical environment and humanoid robot interaction and construct the bottom layer control algorithm for the robot limbs. Firstly, we validate the PID position control on one degree of freedom (D.O.F.) test platform. Moreover, the platform is actuated by a new mechanical/control actuator: SASA (Sensorless Adaptive Stiffness Actuator). SASA consists of differential and released mechanism. In normal operation, control of SASA is as the general servo actuator but the preload of SASA is limited by sensing spring adjustment. While sensing spring detects the unexpected reaction force and deflects under the threshold, which will trigger the released mechanism, the transmission torque will be immediately cut off and only light weight mechanism collides with the environment or human. The experimental results show that the SASA reduces the static interaction force and dynamic collision, which is measured by head injury criterion (HIC). The motion speed of test platform is controlled by an adaptive impedance generator. This generator can set speed and speed range of the test platform to increase operation efficiency and keep the safety characteristics. In addition, integral gain is added to eliminate the steady state error but this controller is not good for safety. The reaction torque observer reduces the ripple of torque and smoothes the output of velocity. The derivative gain decreases overshoot and resonance when the mechanism is low damping system, such as low reduction gear ratio. Afterwards, we design a humanoid robot with 32 D.O.F. which are actuated by direct current (DC) brushless motors with Harmonic Drives. Since the material of robot structure is aluminum and it uses sheet metal working, the robot is quite light and firm. The communication of the robot based on CAN Open is better and faster than RS232. Moreover, 6-axis force sensor and gyro sensor is mounted on the feet and trunk, respectively. In control algorithm, this thesis focuses on PID position and impedance control for the robot’s leg. The control algorithm comprises PID position and impedance control. PID position control is simpler modeling and greater efficiency than impedance control. Therefore, when the leg no contacts the environment, PID position control is suitable for control the leg. We combine Genetic algorithm and Nelder Mead simplex method (GA and NM) to determine controller parameters of PID position control. On the other hand, impedance control is implemented to intensively protect the leg and environment when the leg contacts the ground. Since PID position control and impedance control methods have some complementary property, we combine the two methods and switch the control law according the priority of position tracking and stiffness demand. To verify the control algorithm, we construct one leg physical model and fix the waist of the leg in the ground. The simulation results show that we can determine the good PID parameters for position tracking under PID position control according to the GA and NM method. In addition, when the leg contacts the ground, the interaction force under the impedance control is smaller than PID position control and impedance gain is easier determination than PID position control. Finally, we design walking pattern combined the control algorithm to static walking simulation. When the single support phase, swing leg and support leg is controlled by PID position control and impedance control, respectively. The simulation results show that swing leg can well track the joint trajectories and support leg can maintain the needed stiffness with environment and decrease the output torque which is induced by following error. In the future, we combine the control algorithm and ZMP based stabilization controller to guarantee dynamic walking stable.

並列關鍵字

SASA ； Adaptive impedance generator ； PID Position and Impedance Control ； HIC (Head Injury Criterion) ； Genetic Algorithm ； Nelder Mead Simplex Method

參考文獻

[1] D. W. Robinson, “Design and Analysis of Series Elasticity in Closed-loop Actuator Force Control”, PhD thesis, Massachusetts Institute of Technology, June 2006.

[2] G. Wyeth, “Demonstrating the Safety and Performance of a Velocity Sourced Series Elastic Actuator”, 2008 IEEE International Conference on Robotics and Automation, Pasadena, CA, USA, May 19-23, 2008

[4] K. Hashimoto, Y. Suguhara, H. Sunazuka, C. Tanaka, A. Ohta, M. Kawase, H. Lim, and A. Takanishi, “Biped landing pattern modification method with nonlinear compliance control”, IEEE International Conference Robotics and Automation, pp.1213-1218, 2006.

[5] J. H. Park, “Impedance control for biped robot locomotion”, IEEE Transactions on Robotics and Automation, 17(6):870–82, December 2001.

[7] H. O. Lim, S. Setiawan, and A. Takanishi, “Position-based impedance control of a biped humanoid robot”, Advanced Robotics, vol. 18, no. 4, pp. 415–435, 2004.

國際替代計量

基於位置與匹配控制之類人形機器人系統

主題瀏覽