Heuristics and Policies for Online Pickup and Delivery Problems

A courier company approached NICTA to help develop a decision support tool for managing their daily pickup and delivery problem. They take orders from customers throughout the day and have to allocate one of the vehicles of their fleet to best meet the demand. This decision takes into account a number of hard (e.g. a vehicle must have capacity available for a parcel to be transported, a parcel is picked up and delivered by the same vehicle, a vehicle must wait if it arrives earlier then requested) and soft (e.g. a delivery that occurs outside a given time interval is penalized) constraints. The system we implemented to manage this Online Pickup and Delivery Problem (OPDP) selects an appropriate heuristic to decide the allocation of parcels to vehicles based on previously collected real data from the company. To verify the system working process we are proposing a novel combination of machine learning and optimisation where optimisation is used post hoc to measure the quality of these decisions. The results obtained show that the online schedules are only slightly worse than schedules produced ”offline” and even outperform these for some periods of the day. We conclude by summarizing our efforts and several future improvements.