面向工业机器人装配作业的演示编程关键技术与方法

本项目面向柔性制造行业存在大量的装配工序和转型升级的发展需求，针对目前编程复杂性对工业机器人装配应用发展的约束，研究工业装配演示编程问题。围绕目前工业装配演示编程在高层任务理解和底层执行控制面临的挑战，重点研究人工技能提取理解和机器人执行优化方面的关键理论方法和技术，包括装配作业的空间知识表示与推理方法、装配操作语义理解方法、机器人执行程序的生成与优化，重点解决不确定观测下装配关系和装配位姿的同时推理与估计、连续演示下装配操作显式语义的准确分割与理解、多次演示下装配操作隐式语义的学习与应用、机器人末端执行力位协同控制的学习与优化等关键科学性问题，在此基础上研发人工技能信息采集子系统和工业装配演示编程集成系统，并通过在浙江省特色优势行业小家电和五金产品装配上的演示应用，验证方法和技术的有效性。通过研究，推动机器人演示编程理论与技术研究的突破和在工业领域的应用，并带动我国机器换人的持续发展。

In the new era of industry, the critical demand of upgrading the flexible manufacturingin systems grows significantly. However, the industrial robot, as an important role in this trend, its further application is constrained by the high complexity of programming, especially for the assembly operation. To deal with this difficulty, programming by demonstration (PbD) becomes a promising method to teach robots. Toward the development of PbD system in industrial assembly, this project sets to study both the high level task understanding and low level control, including spatial knowledge representation and inference for assembly tasks, semantic understanding of assembly operations, as well as robot execution program generation and optimization. In the project, several methods are proposed to solve the crucial problems involving joint estimation of pose and assembly relations in noisy observations, segmentation and recognition of continuous unstructured demonstration, inference of implicit assembly operation semantics, and hybrid position/force control and optimization. Based on these techniques and the development of a demostration information collection subsystem, an integrated system for PbD in industrial assembly will be implemented. Its effectiveness will be validated on the real assembly tasks such as small appliance and hardware products, which are the competitive industry in Zhejiang Province. By the research conducted in this project, the breakthrough of PbD theory and the further extension of PbD application can be expected, finally leading to the substantial development in robot assembly line.