SIA OpenIR  > 其他
四足机器人动态行走控制方法研究
Alternative TitleDynamic locomotion control for quadrupedal robots
刘明敏
Department其他
Thesis Advisor曲道奎 ; 徐方
Keyword四足机器人 全身运动控制 运动发散分量 重心动力学
Pages114页
Degree Discipline机械电子工程
Degree Name博士
2021-08-21
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract四足机器人具有非连续地面支撑的运动特点,能够跨越崎岖复杂的地形,具备在非结构化环境中的应用潜力。本研究旨在解决四足机器人运动规划及控制问题,提出了一种基于运动发散分量(DCM)方法论的分层优化运动规划及控制框架,该框架可利用模型对系统进行预测,根据预测结果进行优化来产生当前控制运动的控制参数。应用此框架,实现了四足机器人在复杂状态下的稳定运动。运动规划及控制框架分为三层,这些层次由上到下,预测时域逐级递减,模型逐级复杂,控制周期逐级递减。第一层为步态规划层:将四足机器人抽象成一个线性倒立摆模型,在给定四足机器人落脚点的情况下规划出线性倒立摆模型的运动轨迹。第二层为模型预测控制:根据步态规划给定的四足机器人质心的运动轨迹,在宽松约束条件下最优控制能量的引导,利用MPC的滚动优化策略,求解在短期内轨迹跟踪最优的基座控制参数(足底力)的期望值。第三层为全身动力学控制器:以减小加速度和足底力指令期望值的跟踪误差为优化目标,在满足一系列约束的情况下,在每个控制周期内用反馈信息实时求解最优的控制参数。本文的主要研究内容包括:针对浮动基座多支链四足机器人动力学建模复杂的问题,提出了一种基于李群李代数形式的牛顿欧拉方程的四足机器人动力学方程封闭解求解方法。将四足机器人拆分为四条支链,分别建立四条支链的动力学方程封闭解,然后整合出浮动基座四足动力学方程的封闭解。将其逆向动力学求解过程描述成优化问题,通过构建二次规划求解出最优的控制能量去跟踪控制轨迹。为了使四足机器人在出现较大的轨迹跟踪误差时仍然可以稳定运动,提出了一种基于DCM的在线步态规划方法。扩展了虚拟腿的概念,将四足机器人抽象成三维线性倒立摆模型(LIPM),根据离线规划的落脚点,应用DCM方法论递推出保持DCM有界的参考轨迹;运用宽松初始状态模型预测控制在满足一系列输入约束的情况下在线优化出可快速收敛到参考轨迹上的落脚点以及状态轨迹。针对四足机器人运动控制问题,本研究提出了一种基于DCM方法论的模型预测控制与基于重心动力学的全身运动控制相结合的控制方法。全身运动控制使用全尺寸动力学模型补偿模型预测中模型简化的影响,保证系统稳定;而模型预测控制补充了全身运动控制只能考虑当前状态的局限性,增强了系统的鲁棒性。本章分别提出了两种模型预测控制器,角动量控制器与足底力控制器。角动量模型预测控制与全身运动控制相结合实现了四足机器人双点足着地(欠驱动)平衡控制,足底力模型预测控制的使用增强了四足机器人的运动特性。最后,搭建了四足机器人物理样机,对本文所提出的方法进行综合验证。通过四足机器人姿态变换实验,对角小跑实验和扰动恢复实验验证了本文所提出的浮动基座动力学建模及求解方法、四足机器人步态规划以及基于重心动力学的全身运动控制算法的有效性。
Other AbstractQuadruped robot, which has the characteristics of discontinuous ground support, can cross rugged and complex terrain, and has the application potential in unstructured environment. This research aims to solve the problem of quadruped robot motion planning and control, and proposes a hierarchical optimization motion planning and control framework based on DCM methodology, which can use the model to predict the system and optimize according to the prediction results to generate the best control parameters of current motion. With this framework, the stable motion of quadruped robot under complex conditions is realized. The framework of motion planning and control is divided into three levels. The prediction time domain is gradually decreasing, the model is gradually complicated, and the control period is decreasing. The first level is gait planning: the quadruped robot is abstracted into a linear inverted pendulum model, and the trajectory of the linear inverted pendulum model is planned under the condition of given quadruped robot's footprint. The second level is model predictive control: according to the motion trajectory of the quadruped robot given by gait planning, under the guidance of optimal control energy and loose constraints, the rolling optimization strategy of MPC is used to solve the expected value of the optimal base control parameter (ground reaction force) in short-term trajectory tracking. The third level is the whole body dynamics control: the optimization objective is to reduce the tracking error of the expected value of acceleration and ground reaction force command, under the condition of satisfying a series of constraints, and the optimal control parameters are solved in real time by feedback information in each control cycle. The main contents of this paper include: To solve the complex dynamic modeling problem of multi-branched quadruped robot with floating base, a closed-loop solution method of quadruped robot dynamic equations based on Lie group Lie algebraic form of Newton Euler equation is proposed. The quadruped robot is divided into four branches, and the closed solutions of the dynamic equations of the four branches are established respectively, and then the closed solutions of quadruped robot dynamic equations are integrated. The inverse dynamics solution is described as the optimal control problem, and the optimal control energy is solved by constructing a quadratic programming to track the control trajectory. In order to make the quadruped robot to move stably even when there is a large trajectory tracking error, an online gait planning method based on divergence component of motion(DCM) is proposed. The concept of virtual legs is extended, and the quadruped robot is abstracted into a three-dimensional linear inverted pendulum model (LIPM). According to the footprint of offline planning, the DCM methodology is used to provide the reference trajectory to keep the DCM bounded; the loose initial state model predictive control is used to optimize the desired footprint and state trajectory, which can quickly converge to the reference footprint and state trajectory, under a series of input constraints. For the motion control problem of quadruped robot, this research proposes a control method that combines model predictive control based on DCM methodology and whole body motion control based on centroidal dynamics. Whole-body motion control uses a full-scale dynamics model to compensate for the influence of model simplification in model prediction, and to ensure system stability; while model predictive control supplements the limitations of whole-body motion control that can only consider the current state, enhancing the robustness of the system. This chapter proposes two model predictive controllers, the angular momentum controller and the plantar force controller. The combination of angular momentum model predictive control and whole body motion control realizes the dual-point foot landing (under-actuation) balance control of the quadruped robot. The use of the plantar force model predictive control enhances the motion characteristics of the quadruped robot. Finally, a physical prototype of a quadruped robot was built to comprehensively verify the method proposed in this paper. Through the quadruped robot attitude transformation experiment, trot experiment and push recovery experiment, the effectiveness of the floating base dynamic modelling and solution method, the quadruped robot gait planning and the whole body motion control algorithm based on centroidal dynamics are verified.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/29403
Collection其他
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
刘明敏. 四足机器人动态行走控制方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2021.
Files in This Item:
File Name/Size DocType Version Access License
四足机器人动态行走控制方法研究.pdf(26022KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘明敏]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘明敏]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘明敏]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.