Wednesday, October 15, 2014, 10:30 - 12:00
Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany
A motion simulator is a device designed to provide the human subject with a realistic feeling of being inside a moving vehicle, such as a car or an airplane. Humans perceive self-motion mainly via visual and vestibular systems. While current computer graphics technology can provide very realistic visual cues, realism of inertial cues is strictly limited by physical constraints of the motion system, such as maximum positions, velocities and accelerations of the mechanical axes. The challenge is to control the motion system is such a way that the difference between its output (accelerations and rotational velocities) and the corresponding values in a real vehicle is minimized, while satisfying the motion system constraints. We are currently solving this problem by using a large-scale optimization algorithm to find motion system input for a pre-defined reference output. This approach shows very good results, but is extremely slow. We aim at improving the performance by using more efficient software implementation and/or smarter algorithms. Our final goal is to make our approach to work in a real-time scenario, where a human subject is actively controlling the simulated vehicle. We want to implement a non-linear model-predictive controller that would be fast enough for this purpose.