Multiple Heuristics with Reinforcement Learning to Solve the Safe Shortest Path Problem in a Warehouse

Intelligent vehicles, provided with an ability to move with some level of autonomy, recently became a hot spot in the mobility field. Still, determining what can be exactly done with new generations of autonomous or semi-autonomous vehicles able to follow their way without being physically tied to any kind of track (cable, rail, \(\ldots \) ) remains an issue. We focus here on the top level of hierarchical decision level (distributes and schedules Pick up and Delivery tasks) and deal with the problem which consists in inserting an additional vehicle into an already working fleet and routing it while introducing a time-dependent estimation of the risk induced by the traversal of an arc at a given time. We propose a model and design a bi-level heuristic with dynamic programming and an A*-like heuristic along with three methods to generate speed functions which can all rely on a reinforcement learning scheme to route and schedule this vehicle.