【作者】 王旭阳；

【导师】 吕恬生；

【作者基本信息】 上海交通大学 ， 机械电子工程， 2008， 博士

【摘要】 非规则运动是指参与运动的自由度数目较多、约束条件复杂且运动过程具有非周期性和随机性等特点的运动。它是类人机器人研究领域中的一个前沿应用课题,在康复医疗、家庭服务、星际探险、体育娱乐等领域有着极其广泛的应用前景。相对周期性的步行运动而言,类人机器人实现非规则运动所遇到的最大挑战在于其适应性,包括任务的适应性和环境的适应性。因此,如何面对给定任务和环境而保持或提高机器人的运动能力显得至关重要,其关键就是如何进行运动规划和控制。类人机器人是一个多变量、强耦合、非线性和变结构的复杂动力学系统,其变结构的不稳定性和产生稳定运动所需解决的动态平衡问题,对于控制理论和动力学问题的研究来说具有很大的挑战性。本文以提高类人机器人的运动能力为出发点,对类人机器人非规则运动的动力学建模、运动规划及抗干扰控制方法等关键问题进行了研究,提出了一些新的方法并做了大量的仿真和实验验证。论文主要工作及贡献如下:1)本文针对类人机器人非规则运动的特点,分析了非规则运动在类人机器人动力学建模中所存在的主要问题及其解决方法。介绍了利用Kane方程对非规则运动中简单非完整约束问题进行求解的主要过程,给出了非完整偏速度和偏角速度以及非完整广义主动力和惯性力的推导过程,并说明了非完整系统与完整系统在建模过程中的联系与区别。类人机器人运动学和动力学方程的建立,不仅可以用来计算关节运动的驱动力矩,还为步态和非步态运动规划提供了重要依据。2)在非规则运动规划中,本文提出了基于状态转换的运动规划方法,即根据给定任务的关键运动状态为类人机器人生成满足动力学约束要求的、稳定的运动轨迹。该方法通过建立类人机器人运动的状态空间,将连续的运动采用离散的状态来表示,再通过相邻状态之间的转换(即状态转换)来实现各种复杂的非规则运动。其中关键运动状态的获取是状态转换运动规划方法的基本要素之一。为了替代传统的HMCD运动数据获取方法,本文基于遗传算法提出了一种新的状态生成方法,该方法可以在没有任何参考运动数据的条件下,通过自由指定各种约束条件,生成稳定的且满足约束条件要求的机器人运动状态。与HMCD法相比,该方法是一种更为直接、有效的运动状态生成方法,极大地提高了类人机器人非规则运动数据的获取效率。另外在状态转换运动规划中,状态转换过程可以划分为状态内转换和状态间转换,通过状态转换的识别对其加以区分。在状态转换识别方法上,本文提出了接触点集合的概念,对接触状态的变化进行逻辑上的表达;再结合接触点集合的变化情况,通过设定接触力阀值来实现对状态转换的识别。3)根据给定的状态转换轨迹进行局部的跟踪控制,是本文实现类人机器人非规则运动的另一个重要研究内容。本文针对运动中受小幅外力干扰作用的情况,对类人机器人轨迹跟踪的抗干扰控制方法进行了研究。提出了利用逆动力学控制和ZMP平面的加速度映射相结合的抗干扰策略,在对运动轨迹进行精确跟踪的同时保证系统整体的动态平衡性,解决了经典轨迹跟踪控制方法在抗干扰及稳定性保持等方面的不足。4)为验证本文所提出方法的有效性,分别构建了非规则运动规划与抗干扰控制算法的实验平台,详细地介绍了运动生成和运动执行系统的设计与构成。整个系统为本文通过实验验证所提出的运动规划及抗干扰控制算法提供了前提和基础,也是研制类人机器人非规则运动系统的关键实现技术之一。本文有关类人机器人非规则运动的运动规划和抗干扰控制的相关研究,将有助于提高机器人的拟人程度和运动能力,拓展机器人的应用领域,在理论和应用上都具有一定的借鉴作用和参考价值。更多还原

【Abstract】 Non-regular motions are referring to the ones that involve great degrees of freedom (DOFs), complex constraint conditions, and characteristics of non-periodicity and randomness. As a front research topic in the humanoid robot field, they have well application foreground in fields of rehabilitation treatment, house works, interstellar exploration and sports entertainment.Relative to the periodic biped motion, the challenges for non-regular motions consist in the applicability, especially the applicability for tasks and environments. It’s very important for the humanoid robot facing a given task or environment to keep or promote the motion capability. The essential problems are motion planning and motion control. Humanoid robot is a multivariable, strongly coupled, nonlinear and varying structure dynamic system. Its instability due to the varying posture structure and the dynamic stability for a stable movement is very challenging for control theory and dynamics molding.In order to improve the humanoid versatility, this dissertation concentrates on some key problems relate to the non-regular motion, including the dynamic modeling, motion planning and disturbance rejection control. Simulations and experiments have also been performed to test the methods presented in the dissertation. We have mainly done the following research work:1) In this dissertation, some main factors appearing in the modeling process are analyzed according to the characteristics of the non-regular humanoid motion. In addition, the solution for simple nonholonomic constraints appearing in the non-regular motion is presented by using Kane’s method. And the nonholonomic partial velocities and the partial angular velocities as well as the nonholonomic generalized active and inertia forces are derived to illustrate the relationships between the nonholonomic and the holonomic systems. Apparently, in addition to the calculation of the controlling torques, an important support for gait and non-gait planning is also provided by the dynamics model of the robot.2) In the term of non-regular motion planning, a state transition method is presented for humanoid robot, that is to generate dynamically stable trajectories according to the key motion states of the given task. In this method, motion planning is simplified by introducing a state-space to describe the whole motion series. And by means of transitions between the neighboring states, i.e. state transitions, all kinds of complex motions can be realized. Since the original motion data are one of the basic elements for the state transition motion planning, this dissertation proposes a new state generation theory to replace the conventional HMCD method. And a state generator is developed based on genetic algorithm (GA), which enables users to generate various motion states without using any reference motion data. By specifying various types of constraints such as configuration constraints and contact constraints, the state generator can generate the desired motion states that are stable as well as satisfy the constraint conditions. Compared with the HMCD method, our approach provides a more direct way to acquire the non-regular motion data. In addition, the state transition process is classified into intra-transition and inter-transition categories. In order to distinguish these two types of transitions, a state recognition method is introduced. We firstly propose a variable contact set in which the contact points on the robot body are defined to represent the alternation of the contact states logically. Then, by predefining thresholds for the contact forces, the state transition can be well recognized.3) As another research topic, the local tracking control according to the pre-acquired state transition trajectories plays an important role in realizing the non-regular motion for the humanoid robot. In this thesis, a control strategy for the trajectory tracking under external disturbance forces is presented. The method combines the inverse dynamics control with the ZMP-plane acceleration projection, whereby the whole body stability of the robot can be considered simultaneously when tracking the predefined trajectories. With this method, the inapplicability existing in the classical control methods in the terms of disturbance rejection and stability maintenance is solved.4) Finally, in order to verify the proposed methods, experiment platforms for the motion planning and disturbance rejection control are constructed respectively. And the motion generation and the execution system are introduced in detail. The platforms provide the premise and basis for verifying the proposed motion planning and disturbance rejection control algorithms by experiments. And it is also one of the key implementation technologies for the non-regular motion system of the humanoid robot.The contribution of the dissertation consists in improving the human-like versatility and motion capability of the humanoid robot, widening application field and its reference value on the theory and application.更多还原