In behavior planning for autonomous vehicles, other traffic participants have to be considered. This is typically achieved by using a prediction model to estimate their future behavior and by evaluating how it affects the planned actions of the ego vehicle. The inverse problem of how the ego's behavior may affect the traffic situation, is rarely looked at. This paper presents a novel planning approach which treats the whole traffic situation as the system to control. By means of an interaction-aware prediction, the consequences of the ego's actions for other vehicles are estimated. Via a cost function, an optimal outcome for the whole traffic situation is defined. The optimum is approached in an iterative way instead of using sampling-based methods. The planning combines lane change behavior and longitudinal acceleration. To narrow down the continuous solution space, the available actions are expressed by only four variables. The approach was designed for use on controlled-access highways, but the basic concept is transferable to other scenarios. Its usefulness is demonstrated in three interesting scenarios within a third-party traffic simulator.