We propose to handle execution duration overruns (temporal faults) in real-time embedded systems. When a temporal fault occurs, the slack time can be dynamically determined and assigned to the faulty task in order to complete its treatment. This mechanism improves the temporal robustness of real-time systems. We demonstrate that an approximate slack stealer algorithm like the MASS algorithm is a good solution for real-time embedded systems. We validate the feasibility of this approach by an implementation on the Lego Mindstorm NXT platform.