基于传感器信息融合的四足机器人控制

CASIA OpenIR > 毕业生 > 硕士学位论文

	基于传感器信息融合的四足机器人控制
	黄竹慧
	2018-05
学位类型	工学硕士
中文摘要	四足机器人运动灵活、环境适应性强，但在复杂环境中运动时，其控制难度大，需要感知自身状态及环境信息，以实现基于多传感信息融合的行走控制。本文设计了结构紧凑、接口灵活的四足机器人嵌入式控制器，并研究基于足端轨迹规划与中枢模式发生器的四足机器人节律运动控制，通过对机器人自身运动状态的感知，实现四足机器人的稳定行走控制。在实现平稳运动控制的基础上，结合包括足-地接触力信息的传感器信息反馈，进行多传感器信息融合，实现地面基质实时分类。主要研究工作概要如下：（1）设计了基于STM32的嵌入式控制器与对应的传感器信息采集方案。实现了运用STM32读取包括足端力传感器、陀螺仪在内的多种传感器，并在嵌入式控制器中进行初步处理。实现了STM32与电脑上位机的通信，并做出反馈控制。设计了基于C++的控制端，实时反馈各腿足运动相位、本体姿态角等参数，并自动记录数据。（2）融合基于中枢模式发生器（Central Pattern Generator, CPG）的控制算法与基于模型的控制算法，充分利用两种控制算法的优势实现四足机器人的基本步态控制。利用稳定裕度法(Static Margin, SM)与广泛稳定裕度法（Wide Stability Margin, WSM），通过不同步态行走时静态稳定的支撑区域判定和动态稳定判定，规划四足机器人的落足点与位姿，预先将重心位置规划在稳定区域内，实现了稳定行走与步态变换。通过Adams和Matlab-Simulink联合仿真，实现了四足机器人的平稳行走与Stand-Walk-Trot-Walk-Stand的快速步态变换，并通过质心速度波动、以及足端触地情况分析了步态变换的稳定性。（3）基于传感器信息融合技术，利用关节驱动电流、关节角度、足-地接触力和机器人本体位姿数据，经特征提取后，构造并训练神经网络分类器对地面基质进行分类，并在设计的嵌入式控制器中实现了地面基质实时分类。
英文摘要	The quadruped robot, a kind of biomimetic robots, has a bright perspective because of its superiority in locomotion stability, rapid moving velocity and adaptation to rough terrains in natural environment. Appropriate perception of the kinestate of robot and different ground substrates plays a significant role in realizing adaptive quadruped locomotion. Aiming at improving adaptability of the quadruped robot in unknown terrains with different geologies, this thesis focuses on the design and implement of an embedded controller for the quadruped robot, and realizes adaptive locomotion and ground substrates classification. The main contributions of this thesis are summarized as follows. This research proposes the STM32-based embedded controller architecture to obtain and process sensor information, and control the quadruped robot with sensory feedback. Obtained sensor information including foot-ground contact force and gyroscope are preprocessed and transferred to computer. Parameters including the actual phase of the quadruped limb and attitude angle of the torso are shown and recorded on the designed C++ based control site. A gait strategy is proposed for a quadruped robot using the Central Pattern Generator (CPG) and trajectory planning. The CPG is used to generate rhythmic moving pattern during trot gait, and trajectory planning based on the Stability Margin (SM) is employed to implement walk gait as well as the transition between walk gait and trot gait. Especially, a moving sequence based on SM is developed to realize the transition, and the method can complete gait transition along with switches between conventional trajectory planning strategy and the controller with CPG based on the Hopf Oscillator to take advantages of conventional controller as well as CPG. This thesis conducts the simulation of stand-walk-trot-walk-stand transition and illustrates that the ground contact agree well with expectations. The realization of fast gait transition lays the foundation for adaptive locomotion of quadruped robots in different environments. The quadruped robot Biodog realizes multi-sensor information collection while walking on six different ground substrates. This thesis constructs the neural network for ground substrates classification in a supervised manner and trains the network with collected data after calculation and normalization. The real-time classification of ground substrates is realized in the designed embedded controller.
关键词	四足机器人中枢模式发生器步态变换传感器信息融合
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/20933
专题	毕业生_硕士学位论文
作者单位	1.中国科学院自动化研究所 2.中国科学院大学
推荐引用方式 GB/T 7714	黄竹慧. 基于传感器信息融合的四足机器人控制[D]. 北京. 中国科学院研究生院,2018.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
黄竹慧毕业论文(带签字).pdf（6064KB）	学位论文		限制开放	CC BY-NC-SA