This paper addresses the problem of tidying up a living room in a messy condition with a service robot (i.e. domestic mobile manipulator). One of the key issues in completing such a task is how to continuously select the object to grasp and take it to the delivery area, especially when the robot works in constrained and partially observable environments. In this paper, we propose a quantifiable stratification method that allows the robot to find feasible action plans according to different configurations of objects-deposits, in order to smoothly deliver the objects to the target deposits. Specifically, it leverages a finite-state machine obeying the principle of Occam's razor (called O- FSM), which is designed to integrate arbitrary user-defined action plans typically ranging from simple to complex. Instead of considering a sophisticated model for the ever-changing objects-deposits configuration in the tidy-up task, we empower the robot to make simple yet effective decisions based on its current faced configuration under a generalized framework. Through scenario planning and simulation experiments with the explicitly designed test cases based on the real robot and the real competition scene, the effectiveness of our method is illustrated.