The task of 2D object localization prediction, or the estimation of an object's future location and scale in an image, is a developing area of computer vision research. An accurate prediction of an object's future localization has the potential for drastically improving critical decision making systems. In particular, an autonomous driving system's collision prevention system could make better-informed decisions in the presence of accurate localization predictions for nearby objects (i.e. cars, pedestrians, and hazardous obstacles). Improving the accuracy of such localization systems is crucial to passenger / pedestrian safety. This paper presents a novel technique for determining future bounding boxes, representing the size and location of objects - and the predictive uncertainty of both aspects - in a transit setting. We present a simple feed-forward network for robust prediction as a solution of this task, which is able to generate object locality proposals by making use of an object's previous locality information. We evaluate our method against a number of related approaches and demonstrate its benefits for vehicle localization, and different from previous works, we propose to use distribution-based metrics to truly measure the predictive efficiency of the network-regressed uncertainty models.