This paper aims to model a driver assignment problem which considers drivers states and capacity. The problem can be classified as a capacitated vehicle routing problem (CVRP). In the current literature, CVRP is a topic which has the most papers devoted to solving the problem. In this research, a comprehensive cost function is considered, and the driver can always have sufficient amount goods to deliver, and the variety of drivers is enlarged. These three factors are the improvement of the CVRP when compared to the state-of-the art research. We develop a mathematical model using linear programming approach. The complexity of the models makes the analytical solution is difficult to solve the problem. We develop a modified version of genetic algorithm to suit the problem in this research. We apply the model and algorithm to a e-commerce delivery problem of agricultural products. Based on a simulation from an actual dataset, the objective function from the algorithm differs 1.15% from the global optimal solution, which give the algorithm a promising future to be applied in other sets of problem.