Abstract In the future smart factory, the production system will further head in the direction of on-demand production. Products will be assembled one by one based on different specifications of customers. In these recognitions, this paper considers a method for raising productivity of the robot work cell. Under the assumption that the dual-arm robot assembles products in the work cell where one robot is in charge of all steps of assembling the product, we propose a computational algorithm for searching the efficient assembly sequence and work assignment to the robot hands utilizing reinforcement learning. Furthermore, we intend to use past learning results to determine work plans of robots more effectively. The proposed methods can eliminate or decrease the workload of the robot teaching. In addition, they can contribute to shorten the assembly time of products by giving the efficient work plan. In this research, the basic theory for automating the work planning of actual assembled products is considered using a building block model.

中文翻译：

---

利用过去的学习结果搜索产品装配顺序的算法

摘要 在未来的智能工厂中，生产系统将进一步朝着按需生产的方向发展。产品将根据客户的不同规格进行一一组装。在这些认识中，本文考虑了一种提高机器人工作单元生产率的方法。假设双臂机器人在工作单元中组装产品，其中一个机器人负责组装产品的所有步骤，我们提出了一种计算算法，用于利用强化学习搜索机器人手的有效组装顺序和工作分配. 此外，我们打算利用过去的学习结果更有效地确定机器人的工作计划。所提出的方法可以消除或减少机器人教学的工作量。此外，他们可以通过提供有效的工作计划来缩短产品的组装时间。在这项研究中，使用积木模型考虑了自动化实际组装产品的工作计划的基本理论。