双足步行机器人相关翻译

本科毕业论文

外文文献及译文

文献、资料题目：Walking Control algorithm of

Biped Humanoid Robot

文献、资料来源：期刊

文献、资料发表（出版）日期：1999.6.3

院（部）：理学院

专业：光信息科学与技术

班级：光信112

姓名：王若宇

学号：2011121135

指导教师：赵俊卿

翻译日期：2015.5.14

外文文献：

Walking Control algorithm of Biped Humanoid Robot Many studies on biped walking robots have been performed since 1970 [1-4]. During that period, biped walking robots have transformed into biped humanoid robots through the technological development.Furthermore, the biped humanoid robot has become a one of representative research topics in the intelligent robot research society. Many researchers anticipate that the humanoid robot industry will be the industry leader of the 21st century and we eventually enter an era of one robot in every home. The strong focus on biped humanoid robots stems from a long-standing desire for human-like robots.Furthermore, a human-like appearance is desirable for coexistence in a human-robot society. However, while it is not hard to develop a human-like biped robot platform, the realization of stable biped robot walking poses a considerable challenge. This is because of a lack of understanding on how humans walk stably. Furthermore, biped walking is an unstable successive motion of a single support phase.

Early biped walking of robots involved static walking with a very low walking speed [5,6]. The step time was over 10 seconds per step and the balance control strategy was performed through the use of COG (Center Of Gravity). Hereby the projected point of COG onto the ground always falls within the supporting polygon that is made by two feet. During the static walking, the robot can stop the walking motion any time without falling down. The disadvantage of static walking is that the motion is too slow and wide for shifting the COG.

Researchers thus began to focus on dynamic walking of biped robots [7-9]. It is fast walking with a speed of less than 1 second per step. If the dynamic balance can be maintained, dynamic walking is smoother and more active even when using small body motions. However, if the inertial forces generated from the acceleration of the robot body are not suitably controlled, a biped robot easily falls down. In addition, during dynamic walking, a biped robot may falls down from disturbances and cannot stop the walking motion suddenly. Hence, the notion of ZMP (Zero Moment Point)

was introduced in order to control inertial forces [10, 11]. In the stable single support phase, the ZMP is equal to the COP (Center of Pressure) on the sole. The advantage of the ZMP is that it is a point where the center of gravity is projected onto the ground in the static state and a point where the total inertial force composed of the gravitational force and inertial force of mass goes through the ground in the dynamic state. If the ZMP strictly exists within the supporting polygon made by the feet, the robot never falls down. Most research groups have used the ZMP as a walking stability criterion of dynamic biped walking. To this end, the robot is controlled such that the ZMP is maintained within the supporting polygon.

In general, the walking control strategies using the ZMP can be divided into two approaches. First, the robot can be modeled by considering many point masses, the locations of the point masses and the mass moments of inertia of the linkages. The walking pattern is then calculated by solving ZMP dynamics derived from the robot model with a desired ZMP trajectory. During walking, sensory feed back is used to control the robot. Second, the robot is modeled by a simple mathematical model such as an inverted pendulum system, and then the walking pattern is designed based on the limited information of a simple model and experimental hand tuning. During walking, many kinds of online controllers are activated to compensate the walking motion through the use of various sensory feedback data including the ZMP. The first approach can derive a precise walking pattern that satisfies the desired ZMP trajectory, but it is hard to generate the walking pattern in real-time due to the large calculation burden. Further, if the mathematical model is different from the real robot, the performance is diminished. On the contrary, the second approach can easily generate the walking pattern online. However, many kinds of online controllers are needed to compensate the walking pattern in real-time, because the prescribed walking pattern cannot satisfy the desired ZMP trajectory. In addition, this method depends strongly on the sensory feedback, and hence the walking ability is limited to the sensor’s performance and requires considerable experimental hand tuning. The authors have developed biped humanoid robots through the second approach [12, 13]. Specifically, various online controllers are activated and switched in the successive walking cycle.

At present, biped humanoid robot research groups have developed their own robot platforms and dynamic walking control algorithms. For example, ASIMO of HONDA, WABIAN-2 of Waseda University, and HRP-3 of AIST are well known biped humanoid robots. To date, most biped humanoid robots have performed stable dynamic walking on the well prepared flat floors. Studies involving walking on the uneven and inclined floors are still in the early stage [14, 15]. Dynamic walking on an uneven surface is hard to realize because most biped humanoid robots perform hard position control of the joints by using motors and reduction gears and the response times of the actuators and sensors are low due to the reduction gear and sensor noise. Accordingly, it is impossible for the robot to measure the ground conditions instantaneously and it is also impossible for the robot to appropriately respond even if it measures the ground conditions rapidly. On the contrary, the human ankle can rapidly adapt to changing ground conditions. Furthermore, human muscles can contract or relax quickly with smooth motions.

This paper described a dynamic walking control algorithm that considers local and global inclinations of the floor. The authors propose the use of various online controllers to cope with an uneven and inclined floor based on an enhanced version of a previously proposed dynamic walking algorithm [13]. These online controllers are activated successively within a suitable time in a walking cycle. The performance of the algorithm is demonstrated in walking experiments using KHR-2.

This remainder of this paper is organized as follows: In Section 2, KHR-2, the test robot platform, is introduced. In Section 3, the walking pattern generation scheme for walking motion, the online controller design considering floor conditions, and activation of the online controllers during a walking cycle are described. In Section 4, the performance of the proposed walking control algorithm is assessed and demonstrated through walking experiments involving various floor conditions. Finally, Section 5 concludes the paper a discussion and plan for future work.

KHR-2 is a biped humanoid robot developed in 2003 (Fig. 1). It has been utilized as a test robot platform to develop a walking control algorithm for the authors’biped humanoid robots, KHR-3(HUBO) and Albert HUBO [16]. The height, weight, and

total number of degrees of freedom of KHR-2 are 56 kg, 120 cm, and 41 (6 for each leg, 4 for each arm, 7 for each hand, 1 for torso, and 6 for head), respectively. All joint actuators are brushed DC motors with harmonic reduction gears or planetary gears. The authors realized a self-contained system by putting all mechanical and electronic parts into the robot body. Hence, KHR-2 is tele-operated via a wireless LAN (Local Area Network). Electrical circuit boards of joint motor controllers and sensory devices were efficiently designed for minimum energy consumption. Aluminum was used as the body frame material. The thickness and size were also minimized so as to reduce the weight within an allowable range. For human-like appearance, the ratio of each body part corresponds with the human ratio. The degrees of freedom and dimensions are summarized in Table 1.

The control system architecture of KHR-2 is a distributed control system (Fig. 2). The main computer is installed in the torso, and sub-controllers such as joint motor controllers and sensory devices are distributed throughout the body. Communication between the main computer and the sub-controllers is achieved by using a CAN (Controller Area Network) protocol. The specifications and descriptions of the sub-controllers including sensors are presented in Table 2.

In this section, a suitable walking pattern design for dynamic biped walking and an online control technique considering the floor conditions are described. To make the robot walk, a gait trajectory is designed offline. In the biped robotics research field, the gait trajectory, otherwise known as the walking pattern, generates the relative position trajectories of two feet with respect to the pelvis center. Even a well-designed walking pattern cannot prevent the robot from falling down as a result of large upper body motions, vibrations of the body parts, and an uneven floor. Therefore, an online walking control algorithm composed of various online controllers is essential to maintain the dynamic balance in real-time. These online controllers can finely compensate the joints trajectories (Fig. 3).

中文译文：

双足机器人行走控制算法

自1970年以来，人们进行了很多有关两足直立机器人的研究。在这期间，随着科学技术的发展，两足直立机器人开始慢慢像人型机器人转变，人型机器人逐渐成为社会智能机器人的典型代表形式。许多研究人员预期，人型机器人工业将会成为21世纪的领军行业，最终我们每个家庭将进入机器人时代。人们如此执着于类人型机器人的研究，起源于长期以来人们对像人一样的机器的强烈欲望。此外，在人型机器人社会，给机器人赋予人类的外貌也是可取的。然而，在人型机器人发展中最困难的是让机器人平稳的行走，这在机器人研究中是一个巨大的挑战。因为现在人们对人类平稳行走的原理还没有足够的认识。

早期直立机器人是以极缓慢的速度平稳行走，步速大约10s一步，使用COG 来操控机器人的平衡。COG在地面上的投射点通常是选在双脚之间形成的正多边形内部。通过静态行走，机器人能够在摔倒前的任意时刻执行停止和前行的命令。静态行走的缺陷就是，机器人的行动过于缓慢，并且COG所控制的移动范围较大。

因此，研究的焦点开始逐渐转向动态直立机器人研究，致力于使机器人能够进行每步时间都控制在1s之内的快速行走。如果动态平衡能够维持，那么动态行走和完成小幅度的身体动作也可以实现。但是，如果机器人在运动过程中，由于快速运动而产生的惯性力不能得到有效的控制，直立着的机器人就很容易摔倒。此外，在动态运动期间，直立机器人可能会因为电磁干扰而摔倒或者突然不能终止行走状态。因此，我们引入ZMP理念来控制惯性力。在稳定单一支撑上，ZMP是均等分配，而COG只能稳定单一支撑。ZMP的优势在于，ZMP是将聚焦点定位于预期的在地面上的重力中心点和通过动态行走产生的重力点和惯性点的结合点。如果ZMP能够在双脚形成的正多边形支撑命令中严格执行，直立机器人就永远不会摔倒。因此，许多研究小组都把ZMP作为测试动态行走稳定性的标准。

通常，ZMP操控行走有两种方式。第一，机器人能通过考虑大亮点，如大量定位点和惯性力点以形成模式化，然后根据机器人模型ZMP轨迹用ZMP动态学方法计算出其行走模式。在行走过程中，常用到的是后向传感控制。第二，

机器人通过简单的数学模型如钟摆体系建模，行走模式基于简单的模式和指针实验的有限信息确定。在行走过程中，通过运用许多包括ZMP在内的多种二级反馈数据来在线控制以补偿机器人运动缺陷。第一种方法能够确定一个满足预期ZMP轨迹的精密行走模式，但是由于在实际操作中，数据计算负担极大，所以这一模式的生成有些困难。并且，如果实际操作中机器人的数学模型建模困难，将会使其性能大打折扣。另外，许多在线控制模式在实际运行中要进行行走补偿，因为的很多行走操作并不能令人满意的得到预期ZMP轨迹。此外，该方法基于有效的二级反馈，因此行走能力受到了传感性能和指针实验要求的限制。

现在，直立类人型机器人研究已经有了自己的研究平台和动力行走控制法。例如，日本本田（HONDA）的阿西莫（ASIMO）、早稻田(Waseda)大学的WABIAN-2、和研究所AIST的HRP-3都是我们所熟知的直立类人型机器人。迄今为止，绝大多数类人型机器人可以实现在平地上执行动态行走命令。在早期，人们也对机器人在不平坦或者倾斜的地面上的行走进行了研究。动态行走在不平的地表行走很难实现，因为大多数使用马达和传动齿轮的类人型机器人很难在不平的地表进行定位，并且执行器和传感器的能够用来反应时间也由于传感噪音和齿轮传动而受到影响下降。因此，对机器人来说，即时测定路面情况不太可能实现，而且，即使机器人能快速测定路面状况，它也无法根据实际情况作出恰当的反应。相反的，人类的踝关节能够快速适应不时改变的路况，并且可以通过肌肉的收缩和放松进行平稳的行走。

我们考虑到当地实际情况以及在地板上模拟地球倾斜轨道对机器人动态行走控制算法进行研究，力图通过增强曾经的运算法则，实现在线控制机器人在不平和倾斜的地面平稳行走的设想。这种在线控制是在使行走周期以一个合适的时间循环，用KHR-2实验进行行走运算演练。

KHR-2直立类人型机器人发明于2003年（图1）。它被用于机器人平台测试，以增强直立人型机器人KUR-3（HUBO）和Albert HUBO的行走控制运算。KHR-2的身高、体重和其它相关数据分别为56kg、120cm和41（每条腿为6，手臂为4，手掌为7，躯干为1，头为6）。所有的制动关节是磨毛DC电动机减速传动。我们将放入机器人体内的机械和电子元件做成独立体系，因此KHR-2可以通过无线LAN（广域网）进行远程操纵。加入穿反控制器和感知设备的电器电路板能

够有效判断出最小能量消耗方案。常用铝合金做机器人的外部框架，厚度在允许范围内尽量薄以减少总重量。面部比例和人类的面部比例保持一致。自由度和尺寸列在表1中。

KHR-2架构控制体系是分散式控制体系（图2）。主要计算机安装在躯干上，副控制器如犯戒电机控制器和传感设备分散分布在身体各部分。在主控制器和副控制器之间的信号交流通过CAN（控制局域网）完成。副控制器和传感器的规格和说明在表2中列出。

我们又研究了考虑地面情况的机器人动态行走模式在线控制技术。要让机器人能够行走，需离线设计步伐轨道。在直立机器人研究领域，步伐轨道不是大家熟悉的行走模式，它是根据在两脚和和骨盆之间进行相关定位实现。即使精心设计过的行走模式，也不能让机器人在身体过于巨大、身体摆幅大，地面不平的条件下百分之百避免摔倒。因此，在多种在线控制算法中最重要的是能使机器人在实际行走中保持动态平衡的算法。这些在线控制最终作为补充算法（图3）。