Source code used by Team Fortaleza in the Amazon Last Mile Routing Research Challenge, supported by the MIT Center for Transportation and Logistics.
Given real data on route sequences generated by drivers, the objective of the challenge was to develop a model which could predict route sequences from data on stops and packages to be delivered. Our proposed solution uses a formulation of the problem as an inverse reinforcement learning problem and applies Bayesian optimization to train parameters of an embedded route sequence prediction algorithm.