Restocking-Based Rollout Policies for the Vehicle Routing Problem with Stochastic Demand and Duration Limits

Abstract

We develop restocking-based rollout policies to make real-time, dynamic routing decisions for the vehicle routing problem with stochastic demand and duration limits. Leveraging dominance results, we develop a computationally tractable method to estimate the value of an optimal restocking policy along a fixed route. Embedding our procedure in rollout algorithms, we show restocking-based rollout out performs a priori-based rollout, demonstrating the value of explicitly considering preemptive capacity replenishment in a rollout approach for dynamic routing. We also demonstrate the effectiveness of basic local search versus more sophisticated mechanisms for the heuristic component of the rollout procedure.

Genetic Programming Hyper-Heuristics with Vehicle Collaboration for Uncertain Capacitated Arc Routing Problems

“…The proposed vehicle collaboration scheme is not restricted to the SCP proposed in this paper, but can be easily extended to other decision making processes such as the rollout algorithm (Goodson et al, 2015) and Monte Carlo tree search (Sabar and Kendall, 2015). Extending the collaborative scheme to other frameworks is an area of research we would like to continue with.…”

Logistics Management

“…Stochastic impacts manifests in stochastic travel times (Ehmke et al 2015), demands (Goodson et al 2014), service times (Larsen et al 2002), and customer requests (Thomas 2007). For a recent overview of stochastic vehicle routing, the interested reader is referred to Pillac et al (2013).…”

“…As described in Goodson et al (2013Goodson et al ( , 2014, at decision epoch k, a post decision rollout algorithm identifies a decision in DðS k Þ that maximizes the sum of the current-period reward plus an estimate of the expected reward-to-go from the corresponding post decision state. First, to efficiently apply the algorithm, we reduce the decision spaceDðS k Þ & D S k ð Þ drawing on cheapest insertion routing.…”

“…RA solves a MDP over a certain number of decision points selecting the best decision for each decision point. Later decisions are made by a base heuristic (Goodson et al 2015). The outcomes of the base heuristic are used in the MDP to evaluate the current decisions.…”

Stochastic last-mile delivery problems with time constraints

“…However in the SDDP, customers are not known in advance whereas in the DVRPSD, all customers are known ahead of time. A survey of the DVRPSD literature is described in Goodson et al (2015) and Goodson et al (2013). The primary focus of this literature review is on DVRPs and DPDPs.…”