A method for selecting routes for items being sent from a given beginning node to a given destination node within a network. Each segment of a trip is evaluated and assigned a negative reward based on the deviation from the ideal length of time to complete that segment. Upon arrival of the item at the destination node, the route is assigned a large positive reward. The system is configured to attempt to maximize rewards at all times by selecting the routes with the lowest likelihood of delay. As more and more trips are completed, the control system may define a policy that selects routes based on their probability of yielding the highest reward.