七自由度乒乓球机器人的视觉检测及击球决策研究

发布时间：2017-12-31 03:34

  本文关键词：七自由度乒乓球机器人的视觉检测及击球决策研究　出处：《哈尔滨工业大学》2015年博士论文　论文类型：学位论文

  更多相关文章： 乒乓球机器人 七自由度机械臂 解析逆运动学 双目视觉 运动模糊 击球决策 支持向量回归

【摘要】：乒乓球运动要求参与者具备快速反应和快速决策的能力。基于这些特点，乒乓球机器人成为研究高速、智能机器人系统的理想实验平台。本文以七自由度乒乓球机器人为背景，进行了三方面的研究工作：七自由度机械臂的运动控制、高速运动球体的视觉跟踪和基于学习的球击决策。针对带关节限位的起自由度机械臂，本文提出了一种改进的基于“臂角”参数法的解析逆运动学求解方案。七自由度机械臂的冗余运动被参数化为“臂角”，即机械臂与参考平面之间的二面角。基于上述参数化方法，关节限位对于冗余参数的限制能够表示为可以解析求解的三角不等式。而后，在每个自运动流型上“臂角”的取值范围可以简明且易于实现的三角运算得到；最后对于取值范围内的每个“臂角”值，即可求得符合关节限位的关节配置。本文采用数学证明和实验验证两种方法证明了上述方法的正确性。另外，本文还采用MATLAB符号运算实现了机械臂的正运动学；并通过对比常用微分逆运动学算法，选择了加权最小范数法作为其自由度机械臂的微分逆运动学算法。针对高速运动球体在图像中引入的“运动模糊”，本文提出了一种基于双目视觉的图像处理流程：首先给予背景去除法从图像中提取球体的对应区域；而后通过最小化图像方向导数的L2范数来估计模糊参数并采用Tichardson-Lucy算法得到去模糊图像；最后采用基于RANSAC的圆拟合算法得到乒乓球的中心位置并进而计算出乒乓球在三维空间的位置和速度。上述方法可以有效减轻运动模糊现象对测量精度的影响，从而实现高速乒乓球的精确视觉跟踪。针对击球决策问题，，本文提出了一种基于支持向量回归的击球策略学习方法。机器人的击球过程被形式化为击球评价函数，该函数以来球状态和击球轨迹参数为输入，以回报值为输出。该函数由ε支持向量回归算法对经验数据集进行泛化二得到。在在线决策过程中，采用多初值拟牛顿法最大化击球评价函数以求解出最优击球轨迹。由于基于学习的击球决策不依赖于物理模型，因此它可以有效避免非建模动态和模型参数误差等因素对击球成功率的影响。本文提出的所以算法都在七自由度乒乓球机器人系统上实现，机械臂轨迹规划实验、视觉跟踪实验和击球等验证了算法的有效性。
[Abstract]:Table tennis requires participants to have rapid response and rapid decision-making ability. Based on these characteristics, the table tennis robot is the research of high speed, an ideal experimental platform for intelligent robot system. Based on the background of table tennis robot with seven degrees of freedom, the study of three aspects: seven degrees of freedom manipulator motion control of high speed ball the visual tracking and learning based on decision. To strike the ball joint limit the degree of freedom manipulator, this paper presents an improved method based on arm angle parameter analytical inverse kinematics solution. Redundant motion of seven degree of freedom manipulator is parameterized as the "arm angle", namely machinery the arm and the reference plane. The dihedral angle based on the parametric method, the joint limit for redundant parameter constraints can be expressed as the triangle inequality can be solved analytically. Then, in each self moving The flow pattern on the "arm angle" range can be simple and easy to realize the triangle operation; each of the last "for the range of the arm angle value can be obtained with joint spacing joint configuration. In this paper, two methods to verify the validity of the method is proved by mathematical proof and experiments. In addition, this paper also the MATLAB symbol operation realizes kinematics of mechanical arm; and through the comparison of common differential inverse kinematics algorithm, the weighted minimum norm method as the differential degree of freedom manipulator inverse kinematics algorithm for high speed motion in the image into the sphere of" motion blur ", this paper proposes an image processing process of binocular vision based on the given background removal method: firstly extracting the corresponding areas of the spheres from the image; and then through the L2 norm minimization direction image derivative to estimate parameters and fuzzy Tichardson-Lucy algorithm is used to get the fuzzy image; finally the center position of RANSAC circle fitting method to get the table tennis and table tennis and then calculate the position and velocity in three-dimensional space. Based on the above method can effectively reduce motion blur on measurement accuracy, accurate visual tracking to realize the high speed for hitting of table tennis. The decision problem, this paper proposes a method of learning strategy based on support vector regression. Ball hitting process of the robot is formalized as a batting evaluation function, this function since the ball state and stroke trajectory parameters as input and return to value as output. The function by epsilon support vector regression algorithm of empirical data sets were obtained two generalization online. In the decision-making process, the initial value of the quasi Newton method to maximize the batting evaluation function to solve the optimal path based on learning. Due to stroke A decision does not depend on the physical model, so it can effectively avoid the influence of non modeling dynamics and parameter error and other factors on the success rate of stroke. The proposed algorithm so in the system of seven degrees of freedom table tennis robot, manipulator trajectory planning experiment, visual tracking experiments and hitting the availability of the algorithm is verified.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP242


本文编号：1357936