下肢外骨骼機器人系統(tǒng)參數(shù)辨識和控制方法研究
發(fā)布時間：2019-02-09 20:45
【摘要】：外骨骼機器人是一種可穿戴式機器人,其在行軍作戰(zhàn)、醫(yī)療康復和民用助力等方面具有極廣泛的應用前景。本文就外骨骼機器人的傳感系統(tǒng)和控制算法,從串聯(lián)彈性驅(qū)動、機器人動力學模型、靈敏度放大控制器的設計和在線強化學習參數(shù)優(yōu)化等角度對其展開研究。為增強穿戴者在佩戴機器人行走過程中的舒適性,提高機器人的仿生特性,本文在傳統(tǒng)外骨骼關節(jié)基礎上進行改進,在關節(jié)驅(qū)動電機和負載之間串聯(lián)了彈性元件,用以減緩運動過程中的沖擊作用并存儲運動能量。先對彈性關節(jié)進行了數(shù)學建模,通過MATLAB仿真分析其動態(tài)跟隨特性以確定合適本系統(tǒng)的彈性體剛度。為最大程度上簡化機器人傳感系統(tǒng),本文采用靈敏度放大控制方法,此控制方法不需要任何檢測人機交互力的傳感器,但其對機器人動力學方程及動力學參數(shù)的準確性提出較高要求。本文使用拉格朗日方程推導機器人動力學方程,為最大程度上保證模型準確性,關節(jié)摩擦力矩和電機轉(zhuǎn)子慣量等因素被統(tǒng)一考慮在模型當中,桿件質(zhì)量和質(zhì)心位置均采用實驗的方法進行辨識。所設計的控制器采用通過實驗辨識出來的機器人動力學參數(shù),在定系數(shù)靈敏度放大控制實驗成功之后,為進一步優(yōu)化機器人隨動效果,本文設計了強化學習在線優(yōu)化靈敏度系數(shù)的算法。即下層仍采用靈敏度放大控制,上層采用DMP軌跡規(guī)劃結合強化學習的在線參數(shù)優(yōu)化算法對靈敏度系數(shù)進行在線優(yōu)化。使用DMP算法對人體運動步態(tài)進行學習并給出預測軌跡,其與實際軌跡的偏差作為Q學習算法的實時獎勵。通過MATLAB仿真驗證了控制算法的穩(wěn)定性。最后搭建外骨骼機器人實驗平臺,在外骨骼機器人實驗系統(tǒng)上對所設計的算法進行驗證,證實了動力學參數(shù)辨識的準確性和在線優(yōu)化靈敏度系數(shù)算法的有效性。
[Abstract]:Exoskeleton robot is a wearable robot, which has a wide application prospect in marching, medical rehabilitation and civil assistance. In this paper, the sensing system and control algorithm of exoskeleton robot are studied from the point of view of series elastic drive, dynamic model of robot, design of sensitivity amplification controller and optimization of on-line reinforcement learning parameters. In order to enhance the comfortableness of the wearer in the walking process of the wearing robot and to improve the bionic characteristics of the robot, this paper improves on the traditional exoskeleton joint and makes a series of elastic elements between the joint driving motor and the load. Used to slow down the impact of motion and store motion energy. Firstly, the elastic joint is modeled by mathematical method, and its dynamic following characteristic is analyzed by MATLAB simulation to determine the stiffness of the elastic body suitable for the system. In order to simplify the robot sensor system to the greatest extent, the sensitivity amplification control method is adopted in this paper. This control method does not need any sensors to detect the human-computer interaction force. However, the accuracy of the dynamic equation and dynamic parameters of the robot is very high. In this paper, the dynamic equations of the robot are derived by using Lagrange equation. In order to ensure the accuracy of the model to the greatest extent, the joint friction moment and the moment of inertia of the motor rotor are considered in the model. The mass and centroid position of the member are identified by the experimental method. The designed controller adopts the robot dynamics parameters identified by experiments. After the experiment of constant coefficient sensitivity amplification control is successful, in order to further optimize the robot follow-up effect, In this paper, an algorithm for on-line optimization of sensitivity coefficient by reinforcement learning is designed. In other words, the lower layer still adopts sensitivity amplification control, and the upper layer optimizes the sensitivity coefficient by DMP trajectory planning and reinforcement learning online parameter optimization algorithm. The DMP algorithm is used to study the human moving gait and the prediction trajectory is given. The deviation from the actual track is the real time reward of Q learning algorithm. The stability of the control algorithm is verified by MATLAB simulation. Finally, the experimental platform of exoskeleton robot is built, and the designed algorithm is verified on the exoskeleton robot experimental system, which verifies the accuracy of dynamic parameter identification and the effectiveness of on-line optimization sensitivity coefficient algorithm.
【學位授予單位】：哈爾濱工業(yè)大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP242
本文編號：2419376
[Abstract]:Exoskeleton robot is a wearable robot, which has a wide application prospect in marching, medical rehabilitation and civil assistance. In this paper, the sensing system and control algorithm of exoskeleton robot are studied from the point of view of series elastic drive, dynamic model of robot, design of sensitivity amplification controller and optimization of on-line reinforcement learning parameters. In order to enhance the comfortableness of the wearer in the walking process of the wearing robot and to improve the bionic characteristics of the robot, this paper improves on the traditional exoskeleton joint and makes a series of elastic elements between the joint driving motor and the load. Used to slow down the impact of motion and store motion energy. Firstly, the elastic joint is modeled by mathematical method, and its dynamic following characteristic is analyzed by MATLAB simulation to determine the stiffness of the elastic body suitable for the system. In order to simplify the robot sensor system to the greatest extent, the sensitivity amplification control method is adopted in this paper. This control method does not need any sensors to detect the human-computer interaction force. However, the accuracy of the dynamic equation and dynamic parameters of the robot is very high. In this paper, the dynamic equations of the robot are derived by using Lagrange equation. In order to ensure the accuracy of the model to the greatest extent, the joint friction moment and the moment of inertia of the motor rotor are considered in the model. The mass and centroid position of the member are identified by the experimental method. The designed controller adopts the robot dynamics parameters identified by experiments. After the experiment of constant coefficient sensitivity amplification control is successful, in order to further optimize the robot follow-up effect, In this paper, an algorithm for on-line optimization of sensitivity coefficient by reinforcement learning is designed. In other words, the lower layer still adopts sensitivity amplification control, and the upper layer optimizes the sensitivity coefficient by DMP trajectory planning and reinforcement learning online parameter optimization algorithm. The DMP algorithm is used to study the human moving gait and the prediction trajectory is given. The deviation from the actual track is the real time reward of Q learning algorithm. The stability of the control algorithm is verified by MATLAB simulation. Finally, the experimental platform of exoskeleton robot is built, and the designed algorithm is verified on the exoskeleton robot experimental system, which verifies the accuracy of dynamic parameter identification and the effectiveness of on-line optimization sensitivity coefficient algorithm.
【學位授予單位】：哈爾濱工業(yè)大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP242
【相似文獻】
相關博士學位論文 前4條
1 張超;下肢助力外骨骼機器人研究[D];哈爾濱工業(yè)大學;2016年
2 陳春杰;基于柔性傳動的助力全身外骨骼機器人系統(tǒng)研究[D];中國科學院大學(中國科學院深圳先進技術研究院);2017年
3 龍億;下肢外骨骼人體運動預測與人機協(xié)調(diào)控制技術研究[D];哈爾濱工業(yè)大學;2017年
4 陳珊;行走下肢液壓增力外骨骼自適應魯棒力控制研究[D];浙江大學;2017年
相關碩士學位論文 前7條
1 叢林;下肢外骨骼機器人系統(tǒng)參數(shù)辨識和控制方法研究[D];哈爾濱工業(yè)大學;2017年
2 唐逵;基于有限元方法的下肢外骨骼結構優(yōu)化設計[D];電子科技大學;2017年
3 馬舜;主被動結合式下肢助力外骨骼機器人研制[D];哈爾濱工業(yè)大學;2017年
4 楊陽;外骨骼電機伺服驅(qū)動系統(tǒng)設計[D];電子科技大學;2017年
5 盛文濤;基于人機耦合系統(tǒng)特性分析的外骨骼單腿參數(shù)辨識方法[D];哈爾濱工業(yè)大學;2017年
6 牟洋;拇指功能康復的外骨骼機器人研究[D];哈爾濱工業(yè)大學;2017年
7 郭寧;上肢神經(jīng)康復外骨骼機器人控制策略研究[D];哈爾濱工業(yè)大學;2017年
，本文編號：2419376
本文鏈接：http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2419376.html
最近更新
教材專著
