Online learning for route planning with on-time arrival reliability

Abstract

Consider a network where travel times on edges are i.i.d. over T rounds with unknown distributions. One wishes to choose departure times and routes sequentially between a given origin-destination pair across T rounds to minimize the expectations of: 1) number of rounds when the travel time exceeds an upper bound, and 2) summation over all rounds of the square of the difference between the given target and actual arrival times. We provide an efficient online learning algorithm for this problem.