A method for searching assembly orders by utilizing reinforcement learning

In recent years, the movement to introduce robots in production systems with the aim of replacing manual labor or combining manual and robot labor has started. In particular, dual-arm robots, which have two robotic arms, are gaining much attention in view of their ability to replace manual laborers. However, planning an efficient work system beforehand and teaching the lean movements to the robots are essential for using them more effectively. In assembly processes, the result of selecting the assembly order greatly influences the productivity of the production process. In such a background, this paper examines a method for determining an assembly order with high work efficiency. Under the assumption that dual-arm robots assemble the products that can be assembled easily (i.e., stacking toy blocks is used here), we propose a computational model for searching a highly efficient assembly order utilizing reinforcement learning. We also consider a method for using the results of previous learning model studies to effectively find solutions for new assembly models. Regression analysis is utilized to transfer the past learning results.