Many Response operations are supposed to be occurred right after major disasters. Fatalities increase in the first periods of disasters. So time is a very critical factor in response operations. In these cases artificial optimization methods can be used to manage response teams. These methods make these teams do their activities in a right time. In this paper an improved vehicle routing problem is proposed based on special time window constraint. This constraint is called semi soft time window that uses survival function. Unless hard or soft time window, Semi soft time window models real time varying fatalities. The proposed model is solved with simulated annealing. The operation of this model is evaluated in two small and big randomly generated networks. The results are compared with both soft and hard time window. This led to this conclusion that the new approach can improve the percent of survived people more about 11.5 percent.