步行机器人综合了精密机械、计算机技术、传感与自动控制以及人工智能、模式识别等多学科的最新成果，是复杂的机电一体化技术，是目前最活跃的研究领域之一。而四足机器人由于其步态灵活、承载能力强的特点，又是步行机器人领域中的研究热点。步态控制是实现四足机器人自主稳定步行的关键与核心技术。对步态控制理论进行深入的研究，在实验中实现四足机器人稳定与快速的行走，这无论是在理论上，还是在工程实践中都具有丰富的内涵与深刻的意义。 本文以一种尺寸较大的四足步行机器人为研究对象，以步态控制理论研究与实验分析为中心，目的为实现机器人稳定快速的自然步态行走。对四足机器人步态控制涉及到的中枢模式发生器（CPG）、有限状态机控制器、分层递阶控制结构以及伺服电机控制、动力学与控制的联合仿真等方面进行了深入的研究。论文的主要贡献如下： 1. 针对CPG振荡器网络参数难以根据目标波形的要求进行整定的难点，以Van Del Pol振荡器的参数分析为基础，参考梯度下降学习法则，提出了一种具有自适应特性的振荡器，使振荡器的参数能以目标波形为学习信号进行自整定，输出具有目标波形周期和振幅的信号，降低了设计过程中对经验的依赖，同时使CPG网络本身也具有了一定的自适应特性。 2. 以Van Del Pol振荡器为基础，根据四足机器人的行走特征，构造出适用于四足机器人步态控制的CPG振荡器网络，使髋关节神经元之间以及髋关节与膝关节神经元之间能保持满足自然步态行走要求的相位同步与锁定关系，且此网络参数易于调整，结构简单，计算量小，便于实时控制。 3. 在Van Del Pol型CPG振荡器网络的基础上，通过整合直观控制经验，提出了一种具有环境自适应性、能模拟自然步态的模糊CPG网络步态控制器；实现了四足机器人在上下坡和不平路面上对外界环境的自适应控制；同时，针对具有Pitch-Pitch步行足的机器人，本文通过对机器人转向时动力学的分析，提出了一种通过调整振荡器网络参数实现方向控制的方法。 4. 针对足底触地信号难以引入CPG振荡器网络进行反馈控制的问题，本文结合有限状态机（FSM）理论，对如何处理足底触地反馈信号，实现自然Walk步态进行了研究；提出了四足机器人的有限状态机步态控制器设计方法并进行了仿真和实验。 5. 对四足机器人的分层递阶控制结构进行了研究；提出了实现四足机器人自然步态行走和行为控制的具体解决方案；在伺服电机控制方面，本文根据四足机器人步行足的不同状态，提出利用不同的电机驱动模式并对各个伺服电机的PID控制参数进行模糊优化，有效地改善了驱动电机的性能，使其能更好地适应机器人步行时的实际工况。
英文摘要
Robotics is a complex technology which integrates with precision machinery, computer technology, cybernetic, artificial intelligence and pattern recognition, and it is one of the most active research areas. Because of the Quadruped robot’s flexibility and high load capability, it has become a hot spot of the legged robots. Gait control is the key technology to make the quadruped robot walk steadily and autonomously. Gait control and experiments are meaningful subjects for making the quadruped robot walk steadily and rapidly both in theory and in engineering. The purpose of the quadruped gait control in this thesis is to make the quadruped robot walk steadily and rapidly in the nature gait. The research related to Central Pattern Generator (CPG), Finite State Machine (FSM) Controller, hierarchical control theory, servo motor control and joint simulation of dynamics and control. The main contributions of this thesis include the following issues: (1) We propose an adaptive Van Del Pol oscillator model based on the analysis of Van Del Pol oscillator to solve the difficulties on Van Del Pol oscillator parameter tuning. This oscillator model is able to learn the characters (amplitude, frequency) of the target signal with gradient descent learning method and output demanded signal. By the CPG model with that adaptive feature, the dependence on human experience is reduced. (2) According to the properties and requirements of the quadruped robot’s stable walking, a kind of Van Del Pol CPG model has been proposed. The knee and hip joints CPG neuron are integrated with phase-lock and phase-synchronization relationship to facilitate the gait control. This method has advantage of simpler structure and lower cost of computing. (3) We present a fuzzy CPG gait controller based on Van Del Pol oscillators. By the CPG controller, the quadruped robot’s walk is adapt to the outdoor environments (such as up-slope, down-slope, uneven ground) with bionic gaits. We also propose a control method to enhance the direction control capability of the quadruped robot with pitch-pitch type legs, and the method utilize the unbalance force when the robot’s different side legs touching ground to change direction by tuning corresponding parameters of the CPG controller. (4) We propose a gait controller based on FSM to solve the problem that CPG controller can not work effectively when the foot touch-down signal is incorporated. We test the FSM controller by simulations and experiments...
修改评论