Tasks controlling a mobile system?s interactions with its environment must be performed reliably and in real-time. Motion control, for example, directs a system?s motions based on information about the environment. As such information is usually incomplete and inaccurate, we developed a solution in which mobile systems perform a distributed fusion of their sensor data. In this paper, we address the scheduling of the local processing tasks of the fusion. The particular problem of the real-time scheduling of these tasks lies in their unpredictable execution times, which do not allow using worst-case values. TAFT (time-aware fault-tolerant) scheduling allows working with expected-case execution times and still achieves a predictable timing behavior. It aborts task instances that would miss their deadline and ensures a timely exception handling. We exploit application-inherent redundancy to tolerate task abortions. We present measurements showing the dynamic nature of the execution times and the reliability of task execution.