毕业生知识库

随着制造业产业升级的推进，制造模式逐步向小批量、多品种、柔性化的智能制造模式转变，而人机协作是适应这种柔性制造生产方式的重要发展趋势。机器人与人需要共享工作环境协同作业，这就使得传统物理隔离的安全防护措施难以为继，因此对于人机协作场景中的协作型机器人提出了新的安全要求。为了保证在动态非结构化环境中人机协作的安全性，机器人需要具备对外界环境的感知能力和自主运动规划能力，以适应复杂柔性作业过程的安全协作需求。在这一背景下，本文以常规工业机器人为研究对象，围绕安全控制系统软硬件设计、立体视觉环境感知、自主避障规划三个方面开展研究工作，目的是构成完整的协作型机器人安全控制系统，实现对机器人的协作化升级改造，使其具备应用于人机协作场景的能力。本文的主要研究工作包括：
首先，对协作型机器人的研究背景进行介绍，对目前机械臂的环境感知和路径规划研究现状进行概述，分析协作型机器人应用中的关键技术。
其次，面向协作型机器人的安全性控制需求，基于软硬件分离的模块化设计思想，开展了协作型机器人安全作业平台系统的设计研究，提出基于深度相机的控制系统硬件设计方案和基于ROS的模块化软件系统方案。通过在常规机械臂系统基础上加装深度相机使其具备三维空间的环境感知能力。针对机器人上层控制器需求自主开发了一种双核双操作系统机器人控制器，能够实现对通用任务和实时任务的兼顾。在此基础上，基于ROS构建了由相机通信模块、环境感知模块、运动规划模块和实时控制模块等构成的模块化软件系统。通过机械臂运动学建模和手眼系统标定，实现机器人环境感知与运动控制的基本系统搭建。
再次，针对非结构化动态环境中机器人的感知问题，采用基于立体视觉的信息处理技术，通过Octomap方法对静态环境的障碍物建图，结合基于深度图裁剪和卡尔曼滤波方法的动态障碍物提取，构建了协作型机器人的在线环境感知系统。在静态环境障碍的感知方面，采用了Octomap方法进行建图，为后续的全局路径规划提供环境依据。对于环境中的动态障碍，利用深度图裁剪的方法对深度图中的背景和机械臂进行滤除，实现对动态障碍的粗提取。在此基础上通过欧式聚类方法去除掉提取结果中的噪点，并利用卡尔曼滤波对动态障碍点云中距离机械臂控制点的最近障碍点状态进行估计和跟踪。
然后，为了解决复杂动态场景中协作型机器人的自主规划和实时避障问题，采用基于RRT_Connect的全局路径规划和基于人工势场的局部避障规划相结合的方式，使协作型机器人在感知到环境信息的基础上能够自主规划出全局可行路径并对动态障碍进行实时局部动态避障。在静态环境的Octomap地图的基础上，基于RRT_Connect在机械臂关节空间进行全局路径规划，并通过贪心法和三次样条插值对路径进行优化，获得机械臂的无碰撞路径。为了让机械臂能够对动态障碍做出快速反应，根据人工势场法在动态障碍点和目标构型点分别设立排斥势场和吸引势场，使机器人在两种势场生成的合速度作用下能够实现对动态障碍物的躲避和向目标构型点的趋近。针对机身连杆的避障问题，提出了基于多控制点的机械臂多连杆避障策略，同时对避障速度添加了路径约束，来避免机械臂避障时脱离全局路径与周围环境发生碰撞。
最后对本文进行总结，并提出下一步所要开展的工作。

With the promotion of the upgrading of the manufacturing industry, the manufacturing mode is gradually changing to the small-batch, multi-variety and flexible intelligent manufacturing mode, and human-robot cooperation is an important development trend to adapt to this flexible manufacturing production mode. However, the traditional way of human-robot isolation cannot meet the security requirements of human-robot Collaboration. To ensure the security of collaboration under a dynamically unstructured environment, the capacity of sensing towards outer surroundings and autonomous motion planning is of great needs. Herein this thesis, the conventional manipulator was selected as the candidate and the research was conducted respectively in the aspect of software & hardware design about control system, stereoscopic environment perception and autonomous obstacle-free path planning. The purpose is to completely establish a safely operating system for collaborative robots. The main contents of this thesis are as follows:
Firstly, the research background of cooperative robots is introduced, the current research status of environmental perception and path planning of manipulator is summarized, and the key technologies of cooperative robots are analyzed.
Secondly, for the safety control requirements of cooperative robots, the design research of cooperative robot safe operation platform system is carried out, and the hardware design scheme of control system based on depth camera and the modular software system scheme based on ROS are proposed. By adding a depth camera to the conventional manipulator system, it is able to perceive the environment. A dual-core dual-operating system robot controller is developed to meet the needs of robot upper controller.On this basis, a modular software system consisting of camera communication module, environment awareness module, motion planning module and real-time control module is constructed based on ROS.Through the kinematics modeling of the robot arm and the calibration of the hand-eye system, the basic system construction of the robot's environment perception and motion control is realized.
Thirdly, in order to solve the problem of 3d perception of robots in unstructured dynamic environment, the information processing technology based on stereo vision is adopted to construct the online perception system of collaborative robots to the environment, where the Octomap method is used to identify and map the obstacles in the static environment and the dynamic obstacles are extracted based on background filtering and kalman filtering. Octomap method is used to construct the static environment obstacle map, which provides the environment basis for the subsequent global path planning. For the dynamic obstacles in the environment, the background and the manipulator in the depth map are cropped by background filtering to realize the rough extraction of the dynamic obstacles. On this basis, the European clustering method is used to remove the noise in the extraction results, and kalman filter is used to estimate and track the state of the nearest obstacle point in the dynamic obstacle point cloud from the control point of the manipulator.
Forthly, in order to solve the problem of autonomous planning and real-time obstacle avoidance for cooperative robots in complex dynamic scenarios, RRT_Connect and artificial potential field method are adopted to enable cooperative robots to autonomously plan a global feasible path and conduct real-time local dynamic obstacle avoidance for dynamic obstacles on the basis of sensing environmental information. Based on the Octomap, the thesis performed a global path planning in the joint space with RRT_Connect, and optimized the path by greedy method and cubic spline interpolation, eventually obtaining an obstacle-free path. In order to enable the manipulator to respond quickly to the dynamic obstacles, the repulsion potential field and the attraction potential field are set at the dynamic obstacle point and the target configuration point respectively according to the artificial potential field method, so that the robot can avoid the dynamic obstacles and approach the target configuration point with the combined velocity generated by the two potential fields. As for the obstacle avoidance of the links, a strategy based on multiple control points is proposed, and path constraint is added to the obstacle avoidance speed to avoid collision between the robot arm and the surrounding environment when it is separated from the global path.
Finally, the thesis is summarized and the work to be carried out next is proposed.