Managing production systems in a dynamic environment is a challenging task. As demand and worker utilization change in a factory, the number of work units must be adjusted, which can interfere with efficient component picking and delivery by transfer robots. If the robots could recognize the type of work unit, they could automatically update their maps of picking delivery points to avoid delays. However, this requires a flexible and autonomous method to recognize work units. In this paper, we propose a new approach, LayoutSLAM++, which estimates work units as layout information from sensor data collected by a robot. This method enables adding work unit information to the conventional object map information by simultaneously performing layout estimation and SLAM, considering object confidence and sensor data errors. Layout information can be extracted even when the same object appears in different work units. We validated the method using 15 environments with various geometric constraints and found that the layout information was estimated with 73.0%f-value. Even on a dataset with low geometrical constraints of the object placement, the layout class corresponding to the type of work unit was estimated with 100% precision. Although the recall of layout constraints estimation was as low as 77.7%, the map construction accuracy was improved by 10.6%, showing that the false negatives in layout estimation did not affect the map construction accuracy. By extracting work units as layout information, it is possible to determine the principal object for pick and delivery even when the object's position changes slightly or things belonging to a chair have different appearances. The proposed method can improve the efficiency of transfer robots in dynamic plant environments.