基于深度学习的多机器人协同装配预测控制研究

The multi-robot cooperative assembly and manufacturing technology is one of the priorities for the development of robots in developed countries, and it is increasingly becoming an enabling technology for 3C manufacturing. The multi-robot manufacturing technology of the 3C industrial production line represents the core competitiveness of the national manufacturing industry, which is the strategic focus of the national economy, people's livelihood, and national security. Recent studies have found that there are many excellent features to solve the problems of overlapping interference, synchronization, multi-task driving, and stability from the perspective of model predictive control. However, there are still some core problems that need to be handled for the multi-robot collaborative agile assembly of 3C manufacturing. This project focuses on assembly of server board in 3C manufacturing, and combines advanced assembly technology to expand the predictive control strategy based on deep learning into collaborative assembly systems, and develops distributed predictive control strategies for multi-robot collaborative assembly systems, self-organization decision-making and control methods. The research contents include: high-precision dynamic modeling, real-time obstacle avoidance and path planning under assembly space constraints, assembly task-driven predictive control methods and their applications in board assembly systems. The project is expected to provide multi-robot collaborative assembly solutions for the server board assembly process, which will further provide new technologies and new methods for 3C smart manufacturing research.

多机器人协同装配与制造技术是世界各发达国家在机器人领域未来优先发展的方向之一，也日益成为3C制造业使能技术。3C产业生产线的多机器人制造技术代表着国家制造业的核心竞争力，是关乎国计民生和国家安全的战略重点。近期研究发现，从模型预测控制角度解决重叠干涉、同步、多任务驱动和稳定性等协同问题具有很多优良特性，然而面向3C制造的多机器人协同敏捷装配仍有一些核心难题亟待解决。本项目以3C制造业中的服务器板件装配为主要对象，结合先进装配技术，将基于深度学习的预测控制策略拓展到协同装配系统中，发展多机器人协同装配系统的分布式预测控制策略及自组织决策与控制方法。研究内容包括：高精度动力学建模、装配空间约束下的实时避障与路径规划、装配任务驱动的预测控制方法及其在板件装配系统中的应用。项目预期将为服务器板件装配过程提供多机器人协同装配解决方案，进而为3C智能制造研究提供新技术和新方法。

本项目已按照申请书研究内容和计划执行，并顺利完成了目标，取得了相关的研究成果，主要包括如下几个方面：（1）依据多种传感器网络信息，建立了适用于服务器主板装配的多关节机器人装配系统，通过对不同装配工序精度的分析，设计了相应的模型和定位系统，增加了装配系统的便利性、通用性和柔顺性。（2）设计了多机器人装配系统实时协同避碰避障策略，提出了基于深度学习的机器人误差补偿与轨迹跟踪算法和对距离决策函数进行建模的数据驱动算法，进而结合多机器人运动优化模型，设计了机器人最优几何路径的在线求解策略，保障了装配过程的连续性和平滑性，以及有效降低了重复定位误差。（3）提出了多机器人协同任务中合作子任务和相互作用子任务的状态空间划分方法，阐明了两个子系统的作用机理，并基于此设计了轨迹自适应补偿的多机器人分布式自适应协同控制框架，降低了装配误差。阐明了自组织运动模式形成机制，提出了针对非线性系统的区间二型模糊逻辑控制器，以及设计了考虑输入约束的双模预测控制方法，解决了实际系统中的不确定性和系统约束问题，并为此类多机器人协同装配中存在的共性问题的解决提供了理论基础。（4）研制了多机器人协同装配示范平台，研发了多传感器融合的多机器人装配控制系统，开发了相应的服务器主板装配示范产线。