非结构化环境三维精密装配的机器人手眼协调方法与应用

Aiming at the technical difficulty of three-dimensional precision assembly under unstructured environment, this project proposes a robotic hand-eye coordination method. By moving the vision system with the robotic manipulator, the workspace of the vision system is significantly enlarged. By proposing fast and effective 3D object pose estimation method, the project will solve the workpiece perception problem under unstructured environment. Furthermore, the project will establish self-calibration method for continuous estimation. With the coordination of hand-eye system, the 3D estimation accuracy will be greatly improved by using the high repeatability of robotic manipulation. Specifically, this project will uses the hierarchical search strategy of Pyramid and the method of matching template interpolation to identify the 3D pose of the workpiece at a single frame. Initial pose estimation result will be analyzed to control industrial manipulator to a new location for the next pose identification. This procedure is beneficial to eliminate the wrong pose estimation results that produced in the inappropriate shooting position. At the same time, since the industrial manipulator can move to any position in the work space precisely, multiple pose estimation results that produced in different positions can be converted into the same coordinate system precisely. Then, statistical methods will be employed to analyze these multiple pose estimation results and their error distributions, which are expressed in the same coordinate system, and to obtain the final high precision estimated pose. By moving and seeing, continuous estimation improvement of the proposed robotic hand-eye coordination method, the project will support the development of 3D precision assembly under unstructured environment.

针对非结构化环境中进行三维精密装配的技术难点，本项目提出基于机器人手眼协调的方法，一方面通过视觉系统和机械臂的随动，扩大视觉系统的工作空间，并建立快速有效的三维位姿识别方法，解决非结构化装配环境下的感知难题；另一方面建立基于连续轨迹识别的自标定方法，通过手眼协调，利用机器人高重复运动精度提升三维识别精度，实现三维精密装配，解决高精度执行的难题。本项目的具体将利用金字塔分层搜索策略以及匹配模板插值的方法粗识别，通过识别结果与机械臂的连续轨迹规划的有机结合，利用连续识别消除由于单次识别位置不合理而引起的识别误差。本项目充分利用机器人的局部化运动精度，将系统对同一物体的多次位姿辨识结果转换到同一坐标系下，结合每次辨识结果的误差分布情况利用统计学方法综合分析得出高精度的最终辨识结果。通过“边走边看”以及“越看越准”的手眼协调方法，为非结构化环境下的三维精密装配提供方法和技术支撑。

本项目针对非结构化环境中进行三维精密装配的技术难点，提出了基于机器人手眼协调的方法，以三维物体位姿的视觉识别与定位方法作为研究重点，建立高精度的机械臂与视觉系统有机结合的三维物体位姿识别与定位系统。通过本项目研究，设计和搭建了示范性手眼系统工业应用测试平台，并且开发了手眼标定算法，为项目研究提供了软硬件支撑条件；针对多种工作场景，提出了多类物体识别和姿态估计算法，构建了测试验证实例，验证了所提出的算法的有效性；通过推导和实验验证，验证了多视角观测下，位姿估计结果的异常值剔除和三维姿态回归算法，可以保障机器人手眼系统工作的鲁棒性；通过结合机器学习和分层模板匹配理论，建立了由粗到细、逐步深入多次识别的高精度位姿识别框架理论体系和方法，实现了多类别工业零件的检测、三维姿态估计和分拣。本项目的研究成果已成功应用于焊接生产线和智能档案机器人中，实现了多类零件的分拣和上下料功能，并满足了分拣效率和焊接定位精度的需求。