Multi-Itinerary Optimization as Cloud Service (Industrial Paper) Alexandru Cristian Microsoft a-alcris@microsoft.com Luke Marshall Microsoft Research luke.marshall@microsoft.com Mihai Negrea Microsoft mihai.negrea@microsoft.com Flavius Stoichescu Microsoft favius.stoichescu@microsoft.com Peiwei Cao Microsoft peiweic@microsoft.com Ishai Menache Microsoft Research ishai@microsoft.com ABSTRACT In this paper, we describe Multi-Itinerary Optimization (MIO) ś a novel Bing maps service that automates the process of building itineraries for multiple agents while optimizing their routes to save travel time or distance. MIO can be used by organizations with a feet of vehicles and drivers, mobile salesforce, or a team of personnel in the feld in order to maximize workforce efciency. MIO accounts for service time windows, duration, and priority, as well as trafc conditions between locations, resulting in challenging algorithmic problems at multiple levels (e.g., calculating travel-time distance matrices at scale, scheduling services for multiple agents). To support an end-to-end cloud service with turnaround times of a few seconds, our algorithm design targets a sweet spot between accuracy and performance. Towards that end, we build a scalable solution based on the ALNS meta-heuristic. Our experiments show that accounting for trafc signifcantly improves solution quality: MIO not only avoids violating time-window constraints, but also completes up to 17% more services compared to trafc-agnostic mechanisms. Further, our solution generates itineraries with better accuracy than both a cutting-edge heuristic (LKH3) and an Integer- Programming based algorithm, with twice and orders-of-magnitude faster running times, respectively. CCS CONCEPTS · Information systems → Geographic information systems; Web services; · Theory of computation → Randomized local search; Integer programming. KEYWORDS route optimization, trafc distance matrix, time-dependent travel ACM Reference Format: Alexandru Cristian, Luke Marshall, Mihai Negrea, Flavius Stoichescu, Peiwei Cao, and Ishai Menache. 2019. Multi-Itinerary Optimization as Cloud Service (Industrial Paper). In 27th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SIGSPATIAL ’19), November 5ś8, 2019, Chicago, IL, USA. ACM, New York, NY, USA, 10 pages. https: //doi.org/10.1145/3347146.3359375 Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. SIGSPATIAL ’19, November 5ś8, 2019, Chicago, IL, USA © 2019 Association for Computing Machinery. ACM ISBN 978-1-4503-6909-1/19/11. . . $15.00 https://doi.org/10.1145/3347146.3359375 1 INTRODUCTION In many businesses, route planning and service dispatch operations are a time-consuming manual process. This manual process rarely fnds efcient solutions, especially solutions that can accommodate trafc, location changes or an increasing number of stops along a route. Additionally, scale is also a challenge: service dispatch planning may involve multiple vehicles that need to be routed between numerous locations over periods of multiple days. The development of large scale internet mapping services, such as Google and Bing Maps, creates an opportunity for solving route planning problems automatically as a cloud service. Large amounts of data regarding geo-locations, travel history, etc. are being stored in enterprise clouds, and can in principle be exploited for deriving customized itineraries for corporate agents. The goal of such au- tomation is to increase operation efciency, by determining these itineraries faster (with less man-in-the-loop) and with better quality compared to manually produced schedules. However, multiple challenges stand in the way of making this vision a reality. First, route planning requires efciently calculating the travel- time matrices between diferent locations. While the problem is well understood for free-fow travel times (i.e., assuming no trafc) [8, 20], producing the trafc-aware (e.g., as a function of time-of- day) travel times on-demand and for any point in time requires careful attention to system scalability. Second, the route planning itself has to account for multiple features ś time-windows, the priority of each location, amount of time spent in each location (e.g., service duration or dwell time), and the predicted trafc between locations. The single agent version with no trafc, time-windows, dwell-times, etc. corresponds to the Traveling Salesman Problem (TSP) which is already NP-hard. Numerous extensions to TSP have been studied in Operations Research and related disciplines under the Vehicle Routing Problem (VRP) [22, 23, 25, 29]. However, the bulk of the work is not readily extensible to account for trafc between locations, especially at a large scale. To address customer requirements, our service must incorporate trafc, and output an optimized schedule within seconds for instances with hundreds of waypoints. In this paper, we describe Multi-Itinerary Optimization (MIO), a recently deployed Bing Maps service, available for public use [4]. The design of MIO tackles the above algorithmic challenges, as well as underlying engineering requirements (e.g., efcient use of cloud resources). In particular, our solution consists of a structured pipeline of advanced algorithms. At the bottom layer, we com- pute travel-time matrices by combining Contraction Hierarchies