Finding the optimal location and rotation for every sensor in a multi-sensor setup is not trivial because there are many possible configurations to compare and the space of all feasible positions for the sensors is non-convex. The configuration includes the position and rotation (yaw, pitch, and roll) of the sensors on a car, it influences the ability of an autonomous driving vehicle to perceive its 3D environment. There are some algorithms that optimize the sensor configuration, but all of them lack important properties: They do not consider the accurate shape of the car, the physical size of sensors already placed, or the influence on error assistance to guarantee a safe perception in defined areas. This article introduces a new method that closes this gap. We expand an existing objective function and describe a new genetic algorithm for the optimization which can be applied to any geometric shape of a vehicle. Our method is demonstrated on a simple box model as well as on an accurate car model. These experiments show how the proposed methods can be used to analyze and improve the sensor configuration for any type of autonomous vehicle.