当前位置: 首页>博士论文>资源详情
智能机器人“人-机-环境”交互及系统研究
中文摘要

近年来,在“机器换人”、“无人工厂”、“工业4.0”的大背景下,关于机器人的“人-机-环境”交互及智能作业系统的研究是智能机器人的研究热点之一。本论文研究智能机器人“人-机-环境”交互及系统。提出一种“人-机-环境”交互的智能机器人推理机制。在该推理框架下,对其中的三维语义地图生成过程中的特征描述子进行简化使其兼顾匹配效率和准确率,对其中的地图匹配算法进行优化实现智能交互。机器人能够面向没有任何编程经验的非专业用户、能够实现人机语音交互、能够自动实时的感知环境、能够实现基于三维情景的自主推理与交互、以及实现得到明确作业期望后的自动作业。为了验证本文提出的推理机制,本文搭建了一个“人-机-环境”交互的验证平台,设计了不同的场景来对本文的算法进行验证,并针对机械臂末端精度低的问题,提出了一种基于IMU(Inertial Measurement Unit)末端姿态精确测量的迭代姿态补偿算法,实现低精度机械臂的高精度控制。 本研究的主要工作如下: (1)提出一种“人-机-环境”交互的智能机器人推理机制,以CBR-BDI (Cased-Based Reasoning-Belief-Desire-Intention)推理机制为基础,以用户的中文语音作为用户期望的输入,实现人机交互;采用Kinect来获取三维场景的点云信息并产生语义地图文件,语义地图文件作为推理机的另外一个输入,实现机器人与环境的交互。 (2)在“人-机-环境”交互的智能机器人推理机制中,为了在三维物体识别中兼顾匹配效率与准确率,本文研究三维语义地图生成过程中的局部特征描述子的简化算法。将二进制简化的思想应用于三维视觉识别中的特征描述子的简化中。本文提出一种基于格雷码的二进制特征描述子简化模型,该模型通过选择不同的简化参数(简化单元L和格雷编码位数N)可以产生不同的简化描述子。为了验证本文提出的简化模型,本文将简化模型应用于当前主流的局部特征描述子SHOT(Signature of Histogram of OrienTations)的简化中,得到一种内存占用率低、匹配效率高的新的局部特征描述子G-SHOT。 (3)为了提高“人-机-环境”交互的推理机制的智能交互能力,提出一种对话与三维情景交互算法。通过基于主题树的案例属性的表示、基于机器人语言的案例解决方案的表示以及对话与三维情景交互算法,实现当用户期望不完整或与场景不匹配时的人机对话与引导;用户反馈的信息用于补充/更正之前的不完整或不匹配的期望,直至产生规范化的完整期望并生成解决方案。 (4)提出一种基于惯性测量单元的末端姿态补偿算法,实现了低精度机械臂的高精度控制。搭建“人-机-环境”交互的系统验证平台,验证了算法的有效性。 本文所研究的“人-机-环境”交互的智能机器人推理机制,让机器人拥有了感知(语音交互、环境交互)、思考(推理)的能力,让人机交互变得更加方便和轻松。 关键词:人-机-环境交互;推理机制;三维特征描述子;智能交互;语义地图生成

英文摘要

Recently, under the background of "using robot to replace human", "fully automatic factory" and "industrial 4.0", the research on human-machine-environment interaction and intelligent systems of robots is one of the research hotspots of intelligent robot. Our research work focus on the study on human-robot-environment interaction and systems of intelligent robot. In particular, a human-robot-environment interactive reasoning mechanism is proposed. In our designed reasoning mechanism, a simplified feature descriptor is proposed and used in the process of 3D semantic map generation to balance its ability between matching efficiency and object recognition accuracy; the map matching algorithm is optimized to realize the intelligent interaction of human-robot-environment. Our proposed mechanism is designed for non-expert users who have not been trained in programming, and our mechanism can communicate with human beings by natural language; can real-time percept the 3D environment automatically; has the ability of reasoning and interaction based on 3D scene information; can achieve automatic operation after getting clear and complete desire from users. In order to verify the effectiveness of our proposed reasoning mechanism, an experiment platform for human-robot-environment interactive reasoning mechanism is established to verify our proposed algorithms. For the low accuracy problem of WUST-ARM, an IMU (Inertial Measurement Unit)-based iterative pose compensation algorithm is proposed to improve the end-efFector pose of low-precision modular manipulator. The main contents of our research work are as follows: (1)A human-robot-environment interactive reasoning mechanism is proposed, which is based on Case-Based Reasoning-Belief-Desire-Intention (CBR-BDI) reasoning mechanism. The human-robot interaction is achieved by Chinese natural language, which is an input of our proposed reasoning mechanism. The robot-environment interaction is achieved by Kinect sensor's 3D object recognition and semantic map generation; the semantic map is another input of our proposed reasoning mechanism. (2)In the process of 3D semantic map generation, in order to balance the matching efficiency and object recognition accuracy, a simplified algorithm of local feature descriptor in 3D semantic map generation is studied. A general Gray code quantized model of binary feature descriptors is proposed. In our proposed model, different simplified descriptor can be generate by change the parameters of L and N (L is the encoding group length; N is the number of Gray code bits). Our proposed method is applied to a state-of-the-art descriptor Signature of Histogram of OrienTations (SHOT) to generate new lower memory consumption and high efficient matching descriptor G-SHOT. (3)A dialogue and 3D scene interaction algorithm is proposed to improve the interactivity of the human-robot-environment interactive reasoning mechanism. With representing the case attribution by topic tree; representing the case solution using robot language; and introducing of the dialogue and 3D scene interaction algorithm, the human-robot-environment interactive reasoning mechanism achieves more efficient interaction, reasoning and guidance. When user's desire is incomplete and/or mismatched with the actual scene, robot will take the initiative to guide users through dialogue, and the user's input information will be used to replenish/correct the user's previous desire. At last, system gets the standard and complete desire of users. (4)An IMU-based iterative pose compensation algorithm is proposed to improve the end-effector pose of low-precision modular manipulator. And an experiment platform for our human-robot-environment interactive system is established to verify the proposed algorithms. The methods of intelligent reasoning of robot which are studied in this dissertation makes our robot has the ability of perception (voice interaction, environmental interaction), thinking (reasoning) and makes human-robot interaction becomes more convenient and easy. Keywords: human-robot-environment interaction; reasoning mechanism; 3D feature descriptor; intelligent interaction; semantic map generatio

作者相关
主题相关
看过该书的人还在看哪些书