Development of a Catadioptric Omnidirectional Camera for the USARSim Environment

The first Urban Search and Rescue operation which involved documented use of robots was at the September 2001 collapse of three towers at the World Trade Center. As researchers rarely have the opportunity to do research in real disaster environments, several standardised simulation methods of USAR situations have been devloped. In this thesis I will discuss USARSim, a 3D simulation environment of mobile USAR robots and environment built on top of a 3D game engine created by Epic Gamesand I will describe the development process of a catadioptric omnidirectional camera simulation model for this environment. I will answer the questions whether it is possible to create this simulation model and how much realism is contained in simulated sensor data and thoroughly discuss the method used to create the simulation model. I will show that indeed, it is possible to create simulation model of a catadioptric omnidirectional camera in the USARSim environment and the in terms of image composition it is possible to extrapolate results obtained in the simulation environment using the simulated omnidirectional camera model. This will demonstrated by performing a two phase self localisation method, employing a colour histogram based landmark detection algorithm in combination with an Extended Kalman Filter, in both a real and simulated environment. I will also show that realism of sensor data in terms colour is heavily dependant on the environment in which the camera operates. I will show that, though there is room for improvement of realism in the simulation model, the catadioptric omnidirectional camera is a valuable addition to the USARSim environment. This thesis is a partial fulfilment of the requirements for the title of Master of Science in the field of Artificial Intelligence in accordance with the curriculum set forth by the faculty of Artificial Intelligence of the University of Amsterdam.