【作者】 叶锦华；

【导师】 李迪；

【作者基本信息】 华南理工大学 ， 机械电子工程， 2013， 博士

【摘要】 非完整轮式移动机器人（wheeled mobile robot,WMR）是典型的多输入多输出耦合欠驱动非线性系统，其运动控制问题极具挑战性。轮式移动机器人大多工作在复杂未知环境之下，容易受到多种不确定性和扰动的综合影响，因此，解决复杂不确定下非完整轮式移动机器人的运动控制问题意义深刻且现实需求迫切。本文研究了轮式机器人包含定位不确定性、参数和非参数不确定性、侧滑和打滑干扰等情形下的运动控制策略，探讨了非完整单链系统的有限时间控制以及力矩受限下轮式移动机器人的动力学控制。主要的研究成果包括：（1）研究了定位不确定的轮式移动机器人路径跟随问题，提出一种基于改进遗传算法优化自适应扩展卡尔曼滤波的全局一致渐进稳定控制器。（2）提出了一类n维不确定非完整单链系统的鲁棒有限时间镇定控制律。通过不连续变换将原系统分解为1阶和n-1阶两个解耦的独立子系统，对1阶子系统采用分段控制策略解决不连续变换引起n-1阶子系统奇异问题，保证控制律的全局性，对n-1阶子系统采用反演（backstepping）设计方法，降低设计复杂度，设计过程基于有限时间Lyapunov理论，保证系统的有限时间稳定。（3）研究了本体动力学模型包含参数和非参数不确定性的轮式移动机器人轨迹跟踪问题，提出基于自适应反演滑模控制的全局渐进稳定饱和控制方案。通过运动学输入-输出非线性反馈和动力学输入变换，建立包含系统总体不确定性项的线性模型，采用一种动态调整机制实现控制输入饱和约束，基于幂次趋近律提高了滑模控制的平滑性和快速性，自适应估计总体不确定性的上界有效削弱了滑模控制的抖振现象。（4）提出了执行器动力学模型包含参数和非参数不确定性的轮式移动机器人轨迹跟踪与镇定统一控制方法。通过backstepping分别设计系统的运动学、本体动力学和执行器动力学控制器，运动学控制器引入了时变控制量，使跟踪误差模型用于镇定控制时不存在奇异，本体和执行器动力学控制器分别采用带鲁棒项的强化学习自适应模糊控制补偿系统的复杂不确定性，采用非线性跟踪-微分器避免了backstepping过程的“计算膨胀”，闭环系统为最终一致有界收敛。（5）针对存在未知侧滑和打滑扰动的轮式移动机器人，提出了基于自适应神经网络扰动观测器的鲁棒H轨迹跟踪与镇定统一控制器。采用横截函数（transform function,TF）方法和标准李群运算，建立与原系统等价的输入输出完全解耦的无奇异全驱动统一控制模型，自适应神经网络扰动观测器实现了对侧滑和打滑扰动的精确估计，H控制器对估计误差进行预定水平抑制，消除了未知侧滑和打滑扰动的影响，保证系统的H控制性能。仿真结果表明了上述方法的有效性，最后，论文还设计了一种轮式移动机器人快速实时半物理仿真实验平台，对文中涉及的部分方法进行了实验研究，进一步验证了有效性。更多还原

【Abstract】 Nonholonomic wheeled mobile robot (WMR) is a typical multiple input multiple outputcoupled underactuated nonlinear system, thus its motion control becomes a very challengingproblem. Moreover, most of the WMR work in a complex unknown environment, making itmore easily affected by a variety of uncertainties and disturbances, Therefore, solving themotion control problems of WMR with complex uncertainties has profound significance andurgent practical needs.In this paper, the motion control strategies for WMR with uncertainty of positioning, un-certainties of parametric and non-parametric, disturbances of skidding and slipping are pro-posed. The finite-time control of nonholonomic single-chain system and the WMR dynamicscontrol with bounded torque are also discussed. The major research findings are as following:(1) The WMR path following problem with positioning uncertainty is studied, a globaluniformly asymptotically stable controller based on adaptive extended kalman filtering opti-mized by improved genetic algorithms is proposed.(2) A robust finite-time stabilization control law for the n-dimensional uncertain nonho-lonomic single-chain system is proposed. By using the input-state-scaling technique, the sys-tem is divided into two decoupled independent subsystems, which are a first-order and an(n-1)-order respectively. In order to solve the singular problem of (n-1)-order subsystem withthe discontinuous transformation, a segmented control strategy is applied in the first-ordersub-system, and the global feature of the control law is ensured. The controller designcomplexity of (n-1)-order subsystem is reduced by using the backstepping design method.Since the control law design process is based on finite-time Lyapunov theory, the finite-timestability of the system is ensured.(3) The trajectory tracking problem for WMR with parametric and non-parametric uncer-tainties of body dynamics model is studied, a globally asymptotically stable saturation controlscheme based on adaptive backstepping sliding mode control is proposed. The linear modelwith system overall uncertainty is established by kinematics input-output nonlinear feedbackand dynamics input transformation, and a dynamic adjustment mechanism is adopted for thecontrol inputs saturation constraint. The smoothness and rapidity of sliding mode control isimproved by using power reaching law, the chattering of sliding mode control is weakened byadaptively estimating the upper boundary of sytem overall uncertainty.(4) A unified control method for trajectory tracking and stabilization of WMR with para-metric and non-parametric uncertainties of actuator dynamics mode is proposed. By using backstepping technique, the controllers of kinematics, body dynamics and actuator dynamicsare designed independently, a time-varying control amount is adopted in the kinematics con-troller, making no singular when the tracking errors model is used for stabilization control, thereinforcement learning adaptive fuzzy control with robust item is applied by the controllers ofbody dynamics and actuator dynamics to compensate the complex uncertainties of the system,and in order to avoid the calculation inflation of backstepping procedure, a class of nonlineartracking-differentiator is used. The closed-loop system is ultimately uniformly bounded.(5) A robust H∞unified controller based on adaptive neural network disturbance observ-er for trajectory tracking and stabilization of the WMR with unknown skidding and slippingdisturbances is proposed. By using the transverse function (TF) and standard Lie computingmethods, the no singular input-output completely decoupled unified control model which isequivalent with the original system is established. The skidding and slipping disturbances areestimated by the adaptive neural network disturbance observer, and the estimation error isinhibited with predetermined level by the H∞controller, eliminating the unknown skiddingand slipping disturbances, ensuring the H∞control performance of the system.The effectiveness of proposed methods are verified by the simulation results, and finally,a WMR fast real semi-physical simulation platform is established, and some of the proposedmethods are experimented, the effectiveness of these methods are further validated.更多还原