仿人机器人高效步行模式生成与高稳定动态行走控制方法研究

发布时间：2017-12-27 04:25

本文关键词：仿人机器人高效步行模式生成与高稳定动态行走控制方法研究　出处：《中国科学技术大学》2017年博士论文　论文类型：学位论文

【摘要】：研究开发仿人机器人的最终目的是让其能够像人类一样在现实环境中活动与工作,而高效的步行模式与高稳定的动态行走控制是仿人机器人在现实环境中活动与工作的最基本要求与前提,且也是仿人机器人研究领域的难点与热点。目前得益于硬件技术的飞速发展,使得各种高效复杂的控制算法以及优化算法得以应用于机器人上,来优化与改进机器人的综合性能,进而让仿人机器人能够更加高效稳定可靠的动态平衡行走。在美国国防部高级计划研究局DARPA举办的世界机器人挑战赛DRC的背景下,且在调研总结国内外仿人机器人研究状况的基础之上,本文着重围绕仿人机器人的步行模式规划以及稳定动态平衡行走控制方面展开深入研究,其主要研究内容与创新点如下:1.提出了基于人类行走特征的高效双足步行模式生成方法。首先,以人类步行运动模式为参照,利用NDI高速光学运动捕捉系统捕获了人类步态数据,分析了人类步态的重要性能特征,提取了与双足步行运动重要性能相关的关键步行参数。构建了一种基于关键步行参数的仿人双足步行运动模式综合算法,来为仿人机器人生成完整的步行模式(启动步,循环步,停止步)。然后,构建了一种仿人步行运动模式参数优化算法来提高双足机器人的能效性,该算法分析与定义了有效稳定区域即有效零力矩点变化区域(AZR),推导出了一种双足机器人步行运动模式可行性与能效性的估算方法。最后,通过仿真与步行实验验证了所提方法的有效性,该方法有效的提高了仿人机器人步行的能效性且一定程度上解决了仿人机器人单质心模型不精确引起的实际步态误差问题。2.提出了基于最优可变零力矩点区域(OAZR)与三维线性倒立摆模型(3D-LIPM)的双足步行模式生成方法。在综合分析固定零力矩点与自然零力矩点(X轴方向可变即仿人机器人前进方向可变)的基础上,发现了零力矩点沿Y轴方向(即仿人机器人侧摆方向)变化对双足行走能效与稳定的影响特性,进而得出并定义了最优可变零力矩点区域(OAZR)。基于3D-LIPM,充分利用OAZR构建了一种兼顾能效与稳定的在线步行模式生成方法,其由步行模式综合算法与步行模式参数优化算法所组成,来为仿人机器人在不同地形上行走生成相应的综合性能最优的步行模式。最后,通过仿真与步行实验验证了该双足步行模式生成方法的有效性。3.开展了基于力/力矩反馈的动态平衡行走控制方法研究。构建了一种基于仿人机器人踝关节处力与力矩反馈的稳定动态平衡行走控制策略,其主要由四个控制器所组成,分别是阻尼控制器、位姿与零力矩点补偿控制器、着地姿态控制器以及着地位置控制器。其中,阻尼控制器是基于一种简单的带有柔顺关节的倒立摆模型设计的,用于提高系统的阻尼性能。位姿与零力矩点补偿控制器通过期望零力矩点与实际零力矩点间的偏差以及双足相对位姿偏差来计算出合适的各关节角度补偿量,进而实现对期望的零力矩点轨迹与双足运动轨迹的跟踪。着地姿态控制器是基于踝关节处的力矩反馈实现的,用于摆动脚无阻碍平滑快速的着地。着地位置控制器是用于调整修改摆动脚在垂直与水平平面的位置来解决摆动脚非期望触地引起的瞬间冲击力问题,进而提高仿人双足机器人行走的稳定性与可靠性。4.提出了基于扩展时域被动的摆动脚软着陆控制方法。首先,考虑到摆动脚着地时的碰撞冲击力与摆动脚的着地速度有关,为了避免或削弱X轴方向即仿人机器人前进方向上的触碰冲击力,合适的摆动脚轨迹应该预先被规划。然后,考虑到摆动脚着地误差的存在,通过六维力传感器实时检测摆动脚的力/力矩信息,利用具有导纳因果关系的扩展时域被动控制方法实时在线轻微修正摆动脚与质心位置来抑制或抵消摆动脚着地瞬间的冲击力,并通过斜坡路面行走与轻度不平整路面行走实验验证了软着陆控制策略的有效性。
[Abstract]:The ultimate goal of research and development of humanoid robot is to enable the same activities in the real environment and work like a human, and efficient walking mode and high stable dynamic walking control is the most basic requirements of activities in the real environment and the work of the humanoid robot and the premise, and is also the focus and difficulty of humanoid robot research field. At present, thanks to the rapid development of hardware technology, all kinds of efficient and complex control algorithms and optimization algorithms can be applied to robots, to optimize and improve the comprehensive performance of robots, and to make humanoid robots more efficient, stable and reliable in dynamic balancing. The background held in the DARPA DARPA world robot challenge DRC, and on the basis of summarizing the domestic and foreign research status in the research of humanoid robot, this paper focuses on the humanoid robot walking mode planning and stable dynamic equilibrium running control aspects of in-depth research, the main research contents and innovations as follows: 1. the characteristics of human walking, biped walking pattern generation method based on. First, taking human walking mode as a reference, human gait data were captured by NDI high-speed optical motion capture system. The important performance characteristics of human gait were analyzed, and the key walking parameters related to the important performance of biped walking were extracted. A synthetic biped walking mode synthesis algorithm based on key walking parameters is built to generate a complete walking mode for human like robots (start step, loop step, stop step). Then, constructed a humanoid walking motion model parameters optimization algorithm to improve the efficiency of the biped robot, and analysis of the definition of effective stable region that is the effective area of the zero moment point change algorithm (AZR), and derives a walking motion mode efficiency and feasibility of the biped robot estimation method. Finally, the effectiveness of the proposed method is verified by simulation and walking experiments. The method effectively improves the energy efficiency of humanoid robot's walking and solves the actual gait error caused by the inaccuracy of the humanoid robot's heart model. 2. the biped walking mode generation method based on the optimal variable zero moment point area (OAZR) and the three-dimensional linear inverted pendulum model (3D-LIPM) is proposed. In the comprehensive analysis of fixed zero moment point and zero moment point (X axis variable that humanoid robot direction variable) on the basis of the discovery of the zero moment point along the Y axis (i.e., humanoid robot side swing direction) of the biped walking efficiency and stability characteristics, and proposes the definition and the optimal variable ZMP region (OAZR). Based on 3D-LIPM, make full use of OAZR to build an online account efficiency and stability of the walking pattern generation method, the walking pattern synthesis algorithm and parameter optimization algorithm composed of walking pattern, for humanoid robot walking pattern generation corresponding optimal comprehensive performance in different terrain. Finally, the validity of the biped walking pattern generation method is verified by simulation and walking experiment. 3. the dynamic equilibrium walking control method based on force / torque feedback is developed. To construct a feedback ankle humanoid robot force and torque balance control strategy based on dynamic stability of walking, which is mainly composed of four controllers, namely, damping controller controller, pose and zero moment point compensation with attitude controller and position controller. The damping controller is designed based on a simple inverted pendulum with flexible joints, which is used to improve the damping performance of the system. The position and zero torque point compensation controller calculates the appropriate joint angle compensation amount through the deviation between the desired zero torque point and the actual zero torque point and the biped relative pose error, and then achieves the desired zero torque point trajectory and biped trajectory tracking. The ground attitude controller is based on the torque feedback at the ankle joint, which is used for the swinging foot without hindering the smooth and fast ground. The ground position controller is used to adjust the position of the modified swing foot in the vertical and horizontal plane to solve the instantaneous impact force caused by the undesired touchdown of the swing foot, so as to improve the walking stability and reliability of the biped robot. 4. a soft landing control method based on extended time domain passive wobble foot is proposed. First of all, considering the impact force of swing foot on the ground is related to the ground speed of swing foot. In order to avoid or weaken the X axis direction, namely the impact impact force of humanoid robot in the forward direction, the appropriate swing foot trajectory should be pre planned. Then, considering the swing foot on the errors of moment information through the real-time detection of the swinging leg six axis force sensor using force / time spread with admittance causality passive control methods of real-time online minor amendment swing foot and centroid position to restrain or offset the swing foot with an instant impact force, and the effectiveness of soft landing the control strategy is verified by the sloping surface and mild uneven road walking experiment.
【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】