【作者】 绳涛；

【导师】 马宏绪；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2009， 博士

【摘要】 仿人机器人具有广阔的应用前景,一直以来受到各方面高度关注。经过近四十年发展,取得了辉煌成就,但就目前发展水平来看,还远未达到人类的期望,一个重要原因就是运动性能较差,主要存在速度低、能耗高等问题。欠驱动两足机器人是为了研究高速动态步行而提出的一种机器人结构。机器人没有脚掌,与地面点接触形成欠驱动系统,这给机器人的稳定控制带来巨大挑战,例如:机器人不能形成稳定域,静止站立十分困难;需要通过不断变换支撑点位置实现动态平衡;需要通过时不变步态实现姿态可控等。但是,欠驱动机器人由于可以充分利用重力和惯性力,其步态具有高速、高效、动作自然等特点,为仿人机器人步态规划与运动控制提供了新的思路。本文主要研究欠驱动两足机器人控制策略及其在仿人机器人上的应用问题,主要内容包括:首先对欠驱动两足机器人的复杂控制系统模型进行研究,分别基于D-H坐标建立机器人运动学模型,基于Lagrange函数建立单脚支撑期的控制系统模型,基于角动量守恒建立状态切换模型,并对欠驱动步态的实现条件进行详细讨论,为后面的研究奠定基础。为实现机器人姿态可控,开展了时不变步态规划和控制方法研究,提出了基于遗传算法的时不变步态规划策略:以状态变量为优化参数,以能耗最小为优化目标,以各种约束为限制条件,将步态规划问题转化为非线性规划问题,同时实现步态的规划和优化。基于复杂控制系统模型,机器人运动控制采用反馈线性化方法,因此对解耦矩阵的可逆条件进行了深入研究,并设计了有限时间稳定控制器。稳定控制是欠驱动两足机器人控制最大的难点,也是本文研究的重点。由于机构特殊,现有的稳定判据并不能对其稳定性进行定义和判定。从Lyapunov稳定性理论出发,将机器人状态分为可行和不可行状态。在此基础上,对欠驱动两足机器人的稳定性进行定义,并提出了一种基于失衡度的稳定判据。失衡度是对机器人偏离平衡状态的一种定量评估,综合考虑了姿态和速度参数,满足稳定判据的基本性质。从失衡度收敛的角度出发,给出了欠驱动步态的稳定条件,并以平面欠驱动机器人为平台对欠驱动步态的特点展开研究。实验表明,欠驱动步态具备渐进稳定性,因此,虽然是一种离线规划步态,但是具有很强的鲁棒性和环境适应能力。3D欠驱动两足机器人稳定控制是本研究的主要目标。3D欠驱动步态不稳定主要是由于侧向与前向运动周期不匹配。生物学和机器人学研究表明,两足步行通常以前向运动为主导,侧向辅助保持稳定。基于这一事实,本文提出了一种3D欠驱动步态稳定控制策略,通过控制步宽调整机器人侧向运动周期,实现侧向对前向周期的自适应跟踪,最终实现3D欠驱动步态的渐进稳定。实验表明,该策略对初始状态误差具有很强的鲁棒性,对地面环境具有一定的适应能力,可实现性较强。欠驱动两足机器人静态稳定站立十分困难,但其直立姿态是可控的,本文基于T-S模型和LQR控制设计直立姿态稳定控制器,实现了3D欠驱动两足机器人双脚支撑直立姿态的渐进稳定控制。实验表明,采用所提控制策略,稳定状态的吸引域明显扩大。本文对欠驱动控制策略在仿人机器人上的应用问题展开研究。仿人机器人与欠驱动两足机器人的主要区别在于脚掌和踝关节力矩。利用主动欠驱动策略对支撑踝关节进行零力矩控制,仿人机器人可以实现高速的欠驱动步行。论文重点研究了支撑踝零力矩控制策略的实现方法,如何利用踝关节力矩提高步态稳定性和收敛速度,如何利用踝关节力矩控制步行速度等问题。机器人总是从立定状态开始行走,而最终回归于立定状态,因此论文探讨和解决了仿人机器人欠驱动步态起步和停止过程的控制问题,实现了仿人机器人欠驱动步态行走的全过程控制。实验表明,基于主动欠驱动控制策略,仿人机器人可以实现欠驱动步态,机器人的运动性能明显提高。最后,总结本文所做工作,分析不足之处,并提出今后的研究重点。更多还原

【Abstract】 For the humanoid robot’s potential application, related researches have aroused much attention since its appearance. In the past forty years, research on humanoid robot has acquired remarkable achievements and lots of research organizations built their prototypes to show and validate related results. But to the developmental level of humanoid technology, it achieves far from the expectation of human being. One of the critical reasons is that movement capability of current humanoid robots with huge mass is very poor, and the biped walking is slow and inefficient. So, underactuated biped robot is brought forward to research high speed and dynamic walking. The robot has no soles and touches ground with points, which brings great challenges to biped and humanoid stability control. For examples, it has no stable region, still standing is almost impossible and its supporting points may keep moving to achieve dynamic stability. But the greatest merit is that the underactuated biped robot can walk naturally with high speed and high efficiency. Therefore, the appearance of underactuated biped robot offers new methods for humanoid motion planning and control. The key of this dissertation covers the control strategy and application of underactuated biped robot, and its content as follows:As the basis of research on underactuated biped robot control, the hybrid control system model has been built. Several models are built respectively: kinematics model by D-H reference frame, control system model of single-supporting by Lagrange function, and states-switching model by angular momentum conservation. Then, the realizable conditions of the underactuated gait are discussed in great detail.The planning and control of the time-invariant gait of underactuated biped robot are researched. The gaits are planned and optimized by Genetic Algorithm, which takes the state variables as the optimized parameters, the minimum energy-expend as the optimized goal, the realizable conditions as the restriction, and then the problem of gait-planning is changed into a nonlinear-planning one. The gait is controlled by feedback linearization, so dynamic singularity states and finite-time stability controller are studied.Stability is the principal goal of biped motion control and it is also the most important in this dissertation. The current stability criterion can not be used directly to estimate the stability of underactuated biped robot for its special configuration. Based on Lyapunov stability theory, states of the robot can be classified into two parts: viable and inviable. Then the stability of biped robot is defined and the Lose Balance Degree (LBD) is taken as the stability criterion of underactuated biped robot. LBD is used to evaluate the degree of current state to balance one and synthesizes the pose and speed of the robot. Stable conditions of underactuated gait are given based on LBD, and the characters of underactuated gait are validated based on planar biped robot. Underactuated gait has the characters of gradually stability, so time-invariant gait has strong robustness and adaptability although it is planned offline.The stability control of 3D underactuated biped robot is the main goal of this dissertation. One of the critical conditions for stable biped walking is that the lateral motion should be in accordance with frontal one, where the frontal motion is the dominative part. Based on this fact, the stability control strategy of 3D underactuated gait is brought out which extends the planar one with step width adjustment to control the lateral walking cycle. By using this strategy, the lateral step cycle can be controlled to follow the frontal one and finally the stability of 3D underactuated gait is controlled. This strategy is robust to initial error and realizable for real robot. The stable still standing is very difficult for underactuated biped robot, but the posture is controllable. Then the stability controller based on T-S model and LQR controller has been designed to realize the asymptotic stability of double-support upright standing.Using above proposed strategies, application on humanoid robot has been studied. The humanoid robot can be look on as an underactuated system by active underactuated method and high-speed dynamic walking can be realized by underactuated gait. Emphases of this part include: how to realize ankle Zero-Moment control, how to improve the stability and convergence speed and how to control the walking speed by the moment adjustment of supporting ankle. Bipedal walking always starts and ends with still standing, so the control of starting and stopping processes are studied and fully underactuated walking process for humanoid robots can be realized.Finally, the main content of this dissertation is summarized and key points of future research are discussed.更多还原