A mobile robot follows a resident and grabs his/her health condition using a Kinect sensor. The 3D environment of the robots working space has a huge impact on the design and operation of the mobile robot in two aspects: (1) it defines movements of the resident; (2) it affects the view and trajectories of the robot. This paper proposes an efficient and lightweight floorplan generator which automatically samples 2D semantic floorplans. With the height function, generated floorplans can be converted to diverse 3D indoor scenes. Secondly, based on a floorplan, movements of the resident during a period can be generated automatically with his/her activity schedule. Generated 3D scenes with resident movements will be used to evaluate how indoor spaces affect the design and operation policies of the mobile watching robot. The diversity of scenes with movements has two benefits: (1) providing massive data to constitute a training set for machine learning algorithms; (2) various scenes can be classified for finding statistical conclusions.