This paper reports the first step in an attempt to examine the relationship between simulators and reality. Whereas some simulation efforts attempt to model physical phenomena in the greatest possible detail, others have been successful by adding pseudo-random noise sufficient to swamp inaccuracies that come from an intentionally simplistic model. For example, the ns network simulator attempts very detailed simulation of network links, queues, and radio propagation. In contrast, the arena robot simulator simulates sensors that are physically dependent on physical phenomena by using a trivial model and adding random noise. arena works well because the noise component of the simulated sensor readings swamps errors due to insufficiencies of the model.
It is not yet entirely clear why such a detailed simulation is required for ns, but not for arena. One possible explanation is that network controllers work in environments where ``sensor error'' is not expected--the environment is sensed with perfect fidelity on an extremely short timescale. Robot controllers, on the other hand, have inherent to their design the assumption that the environment is not perceivable without error. They make control decisions on a much longer timescale, and always with the assumption that the sensors may be lying. arena's method works because controllers robust enough to work in the face of arena's random error are also likely to work when faced with the vagaries of the real world.
Whatever the reason for the difference in simulation philosophy, the dichotomy has become more prominent due to projects that simulate simultaneously in both domains; e.g., robots that are both physical actuators and entities on a network. The purpose of this study is to determine whether or not a network simulated with arena-like simplicity can be used when the network is operating in the domain of a robot controller.