Within the last decade the most important automotive OEMs have created and released the system architecture standard AUTOSAR, and tools to support the development process are widely available. However, the resulting system architecture, which is logically modeled in the Virtual Functional Bus (VFB) and realized by generat- ing a Runtime Environment (RTE) which corresponds to the concrete network of ECUs is static in terms of runtime adaptability. As long as vehicle functions dominate which are executed exclusively on separated ECUs, there is only a limited demand for runtime adaption. However, as soon as several vehicle functions are grouped to run on one ECU or one vehicle function is composed by several independent software components, runtime adaption is getting increasingly interesting regarding energy-efficiency for example. In this paper, a concept of self-adaption for AUTOSAR is outlined on the example of a collision detection and warning system. Its timing correctness during the self-adaption process to fulfill the statically defined system architecture is verified with timed automata and the model checking tool UPPAAL.