【作者】 倪修华；

【导师】 陈维山；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2010， 博士

【摘要】 最初的被动步行机器人是一种能在向下的斜坡上仅靠重力步行的纯被动步行机器人,后来在机器人的踝关节或髋关节处引入主动的驱动与控制,发展成为了可沿平地步行的半被动步行机器人。与传统的主动步行机器人相比,被动步行机器人具有步态自然、能量效率高等优点。被动步行的研究成果除可应用于双足机器人的研制外,还可应用于医疗康复设备和假肢的设计,以解决下肢残疾或截肢病人的康复问题。本文主要对双足被动步行机器人的性能进行了分析以及对一种半被动步行机器人的动力输入方法进行了研究。首先介绍了所研制的考虑髋关节弹簧、髋关节阻尼以及腿质心前后偏移影响的二维双足无膝关节纯被动步行机器人样机,并建立了摆动过程动力学方程和碰撞过程动力学方程,利用数值软件MATLAB对动力学方程进行了仿真求解。采用Newton-Raphson迭代方法求出了稳定的不动点。分别采用在ADAMS软件中建立虚拟样机仿真和图像测量的方法验证了数值仿真的正确性。研究了弹簧刚度、阻尼系数和质心偏移等参数对纯被动步行机器人抗干扰能力、步行速度、步行效率以及分岔特性的影响。使用无量纲步态敏感范数的倒数对机器人抗干扰能力进行评价。实验验证了弹簧刚度和质心偏移对机器人抗干扰能力的影响。在实验中,使机器人在各弹簧刚度参数下和质心偏移参数下沿斜坡向下行走100次,记录下行走到斜坡终点的次数作为抗干扰能力的度量。实验结果表明存在一个大小适中的弹簧刚度使机器人的抗干扰能力最大,存在一个较小的质心向后偏移使机器人的抗干扰能力最大。调整弹簧刚度和质心偏移的方法具有操作简单,提高抗干扰能力显著的优点。研究了弹簧刚度、阻尼系数和质心偏移对机器人抗干扰能力影响的作用机理。提高抗干扰能力是被动步行机器人的一个研究热点,而多参数优化方法是提高抗干扰能力的一个非常有效的方法。以步态敏感范数的倒数值最大化为优化目标,使用遗传算法和模式搜索相结合的方法对弹簧刚度、质心偏移、轴向质心距离和转动惯量4个参数同时进行优化。获得稳定的不动点是求解步态敏感范数的前提。而采用Newton-Raphson迭代搜索不动点时,当所选初值偏离不动点较远时难以收敛,搜索时间和搜索成功与否都依赖于初始值的选取。因此使用BP神经网络对被动步行稳定不动点进行估算,并将估算值作为Newton-Raphson迭代的初值求解稳定不动点。参数变化较小时不动点的变化也较小,利用这一特点,在计算相邻两个样本的不动点以获取训练样本时,只使一个参数发生较小的变化,并将本次使用Newton-Raphson迭代搜索得到的稳定不动点作为下一样本使用Newton-Raphson迭代搜索不动点的初值。使用1000组随机产生的参数对所得神经网络进行测试,结果表明该方法不仅大幅提高了稳定不动点的搜索成功率,而且大幅缩短了搜索时间。将该方法应用于优化过程中稳定不动点的求解。首先利用遗传算法强大的全局优化能力进行优化,并将优化结果作为模式搜索的初值,利用模式搜索强大的局部搜索能力进行优化。通过与样机步态敏感范数的倒数值进行对比,验证了优化算法的有效性。由于目前的半被动双足步行机器人所采用的控制方法大多较为复杂,因此由人类步行的生物力学研究得到启发,提出一种半被动双足步行机器人的动力输入方法,并定量分析了各参数对机器人性能的影响。由机器人摆动腿与地发生碰撞后开始于髋关节处施加方波力矩,作为动力输入。由于力矩在一个步行周期中没有做负功,该机器人具有与人类步行相似的高能量效率。该机器人能够实现沿上坡、下坡和平地步行。通过对机器人动力输入参数的适当调整,可以将原本摔倒的步态或分岔的步态控制到稳定的单周期步态上来,并且与纯被动步行机器人相比,半被动步行机器人可以在一个更大的斜坡角度范围以单周期步态稳定步行。更多还原

【Abstract】 The first passive dynamic walker is a pure passive walker which can perform a stable walking motion down a slight slope powered only by gravity. Powered robots that can walk on a level floor by ankle push-off or hip actuation have also been built based on this concept. Compared with traditional walking robots, passive dynamic walkers have more natural-looking gait and higher energy efficiency. The research results can be used to the design of biped robots, medical rehabilitation devices and prosthetic legs to help rehabilitation for the amputees. The performance of a passive dynamic walker and a power input method for a quasi-passive dynamic walker is studied in this paper.A 2D kneeless biped passive walking prototype with hip spring and offset of center of mass was proposed. The equations of motion of the swing phase and the impact phase were derived, respectively. The equations were solved with MATLAB software by numerical simulation. The stable fixed point was obtained by Newton-Raphson iteration. The numerical simulation is verified by virtual prototype simulation and experiment, respectively.The effects of spring stiffness, damping coefficient and offset of center of mass on disturbance rejection, walking speed, walking efficiency and bifurcation characteristics were studied. Reciprocal of the gait sensitivity norm was used as a measure for disturbance rejection. The dimensionless mechanical cost of transport was used as a measure for walking efficiency. The effects of spring stiffness and offset of center of mass on disturbance rejection are verified by experiments. For experiments, we quantified disturbance rejection by observing 100 passive walking trials down a gentle slope of finite length and recording the fraction of trials which successfully walked to the end. The experiments show that an optimal spring stiffness and an optimal backward offset of center of mass for disturbance rejection exist. The method of adjusting spring stiffness and offset of center of mass has the advantages of easy to implement and improving disturbance rejection significantly. The reasons of the effects of spring stiffness, damping coefficient and offset of center of mass on disturbance rejection were investigated.To improve disturbance rejection is a research focus of passive dynamic walker. Multi-parameter optimization is a very effective way to improve disturbance rejection. The genetic algorithm combined with pattern search method was used to find the maximum reciprocal of the gait sensitivity norm. Four optimal parameters were spring stiffness, offset of center of mass, axial displacement of center of mass and moment of inertia. Obtaining the stable fixed point is the premise of obtaining the gait sensitivity norm. Newton-Raphson method does not converge with improper initial values. The search time and successful search rate are dependent on the initial value. BP neural network was used to estimate the fixed point, and the estimated fixed point was taken as initial value for Newton-Raphson iteration to search the stable fixed point. Small variation of parameters leads to small variation of fixed points. When calculating fixed points of two successive samples, only one parameter had a small variation. The fixed point obtained with Newton-Raphson iteration was used as initial value for the next sample. 1000 groups of parameters were generated randomly to test the performance of the neural network. The results show that this method increases the successful search rate significantly and reduces the search time significantly. This method was used to obtain the stable fixed points for the optimization. The genetic algorithm was used for global optimization, then the optimal parameter was used as initial value for pattern search method to optimize locally. The effectiveness of the optimization is verified by comparing the optimal result with the prototype’s reciprocal of the gait sensitivity norm.The present control methods for quasi-passive walkers are complex. Inspired by biomechanics in human walking, we presented a power input method for the quasi-passive walker and quantitatively studied the effects of parameters on the performance of the quasi-passive walker. After the impact between swing leg and ground, a square wave torque was applied at the hip as power input. Since the torque does not do negative work during the whole walking period, the walker has a high walking efficiency as human walking. The robot can walk up a slope, walk down a slope and walk on flat ground. The falling gait and bifurcation gait can be controlled to stable period-1 gait by properly adjusting power input parameters. The quasi-passive walker can walk with stable period-1 gait in a larger scope of slope angle compared with the pure passive walker.更多还原