In the case of simulators of mobile robots, it is necessary to clarify the similarity between actual and simulated robots in order to improve the efficiency and transparency of simulators as research tools. Because of its real-time performance, USARSim  has been popularly used in studies involving human-robot interactions. However, since USARSim uses approximations and abbreviations in its physical computations, there arc some doubts over its computational accuracy. The objective of this study is to compare the actual and simulated robots in order to investigate whether the simulated environments can be used as a training tool. The authors modeled two types of response robots -IRS Soryu |2] and Kcnaf |3]- and compared the abilities of the simulated robots and the actual robots in navigating obstacles such as steps and trenches. The dimensions such as heights, widths and lengths and total mass of the real robots were accurately modeled in the simulation and the other physical parameters were assigned default values. It was observed that the maximum error in the sizes of the traversable objects was 18 % of the body length of the robot. It was also found out that the angular velocities of the robots were significantly different for the real and simulated robots. The conclusion is that a training system can be built based on USARSim when the simulated environments involve simple obstacles because there are not significant discrepancies between the basic traversing abilities of the actual robots and those of the simulated robots. However, in terms of the dynamic motions of the robots the operators need further trainings using the real robots.