Amazon.com has recently released a publicly accessible copy of the dataset used during the 2021 Last Mile Routing Research Challenge.
This dataset is further described in a short technical article published in Transportation Science.
The technical article can also be used to cite the dataset in any academic publication that wishes to reference the data:
Daniel Merchán, Jatin Arora, Julian Pachon, Karthik Konduri, Matthias Winkenbach, Steven Parks, Joseph Noszek (2022). 2021 Amazon Last Mile Routing Research Challenge: Data Set. Transportation Science 0(0). https://doi.org/10.1287/trsc.2022.1173
In connection with the “Last-Mile Routing Research Challenge” co-hosted by Amazon and the MIT Center for Transportation & Logistics in 2021, we are excited to announce a special issue of Transportation Science focused on machine learning-based approaches to large-scale route planning problems. We invite authors independently of whether they took part in the research challenge to submit research papers that address large and complex planning problems related to vehicle routing with nonconventional, data-driven methods. The deadline to submit papers is January 15, 2022
Technical Proceedings of the Amazon Last-Mile Routing Research Challenge have been published by the MIT Megacities Logistics Lab. The proceedings document, available through MIT DSpace, provides a consolidated overview of the short technical papers finalists submitted to document their methodological approach and its expected performance.
Winners of the Amazon Last-Mile Routing Research Challenge were announced in a live webinar event held on July 30, 2021. Amazon and the MIT Center for Transportation & Logistics created the challenge to engage with a global community of researchers across a range of disciplines, from computer science to business operations to supply chain management, challenging them to build data-driven route optimization models leveraging massive historical route execution data. The three winning teams were awarded prize money totaling $175,000 for their innovative route optimization models.
Winners ($100,000): Professor William Cook, University of Waterloo, Canada; Professor Stephan Held, University of Bonn, Germany; and Professor Emeritus Keld Helsgaun, Roskilde University, Denmark.
Runners Up ($50,000): Xiaotong Guo, Qingyi Wang, and Baichuan Mo, doctoral students at MIT.
3rd Place ($25,000): Professor Okan Arslan, HEC Montréal; and Rasit Abay, PhD student at the University of New South Wales, Australia.
Congratulations to all winners and participating teams for such an excellent challenge. Each model addressed the challenge data in a unique way. The methodological approaches that were chosen by the participants frequently combined traditional exact and heuristic optimization approaches with non-traditional machine learning methods.
Watch the award ceremony here.
00:01-11:24 – Setting the context of the research challenge
11:25 – 18:16 – Outline of the collaboration between MIT CTL and Amazon
18:17 – 20:52 – Review of challenge problem set and criteria
20:53 – 25:52 – Overview of participants, locations, and demographics
25:53 – 29:59 – Winners’ announcements and leaderboard
33:00 – 40:25 – Future publications and conclusion
The research challenge attracted more than two thousand interested participants from around the world. 229 researcher teams formed during the spring of 2021 to independently develop solutions that incorporated driver know-how into route optimization models with the intent that they would outperform traditional optimization approaches. Out of the 48 teams whose models qualified for the final round of the challenge, these teams’ work stood out above the rest.
Author: Douglas Gantenbein. This article originally appeared at Amazon Science
Finding the optimal route between multiple destinations — the traveling salesman problem — is a challenge regularly faced by Amazon’s Last Mile team. Meeting that challenge has meant developing planning software to allow Amazon’s delivery fleet to find the most efficient routes. But what happens when drivers must deviate from those routes? Drivers have access to real-time information — road blockage, congestion, parking, etc —and other knowledge and know-how that existing optimization models don’t capture.
“Despite the tremendous advances in routing optimization over the last decade, there remains an important gap between periodic route planning and real-time route execution,” said Beryl Tomay, Amazon vice president for Last-Mile Delivery.
That’s why in February Amazon collaborated with MIT’s Center for Transportation & Logistics (MIT CTL) to develop a competition that challenged academic teams to train machine learning models to predict the delivery routes chosen by experienced drivers.
On July 30, the winners of that contest, dubbed the Amazon Last Mile Routing Research Challenge, were announced. “The Last Mile routing challenge is a classic problem,” said Daniel Merchan, a senior research scientist on Amazon’s Last Mile team. “Our participants worked for more than three months to come up with innovative, data-driven solutions.”
The winners of the Amazon Last Mile Routing Challenge are, left to right: William Cook, professor of combinatorics and optimization at the University of Waterloo; Stephan Held, associate professor with the Research Institute for Discrete Mathematics at the University of Bonn, Germany; and Keld Helsgaun, associate professor emeritus in computer science at Roskilde University, Denmark.
Team Passing Through, comprising scholars from three separate universities, took top honors, winning $100,000. The team’s members are:
William Cook, professor of combinatorics and optimization at the University of Waterloo, Canada; Stephan Held, associate professor with the Research Institute for Discrete Mathematics at the University of Bonn, Germany; and Keld Helsgaun, associate professor emeritus in computer science at Roskilde University, Denmark.
“The challenge was a huge contribution to the research community, providing a massive collection of test instances,” the team said in a statement provided to Amazon Science. “In a single post, Amazon made publicly available more real-world examples of the traveling salesman problem than had been collected in total over the past 70 years.”
Team Permission Denied won second place and featured, left to right, Xiaotong Guo, Qingyi Wang, and Baichuan Mo.
The $50,000 second-place prize went to Team Permission Denied, comprising a trio of MIT Ph.D. students — Xiaotong Guo, Qingyi Wang, and Baichuan Mo — while Team Sky is the Limit, comprising Okan Arslan, assistant professor at HEC Montreal, and Rasit Abay, a Ph.D. student at the University of New South Wales Canberra, won the $25,000 third-place prize.
More than 220 teams participated in the competition, with 45 competing in the final round. Overall, the teams represented 71 different universities and 22 countries. Entrants ranged in academic level from undergraduate to retired faculty.
Team Sky is the Limit, comprised, left, Okan Arslan, assistant professor at HEC Montreal, and Rasit Abay, a Ph.D. student at the University of New South Wales Canberra.
Entrants were given 6,100 historical route records from five areas across the United States to use as a baseline for their project. They also were given more than 3,000 traces of driver-determined routes. Both datasets included driver knowledge. The initial dataset was used for training and testing the model, while the second dataset was utilized for evaluation using both sources of information. Contestants endeavored to build models that could identify and predict drivers’ deviations from routes computed in the traditional manner.
Participants utilized a variety of approaches, including conventional optimization models (some of them enhanced with machine learning components), and wrote short technical papers explaining their approach.
Last Mile Challenge finalists will have an opportunity to publish their research in Transportation Science.
Call for papers A special issue of Transportation Science will be published in conjunction with the Last Mile Challenge. Karthik Konduri, a senior research scientist on the Last Mile team, and Julian Pachon, director and chief scientist for the Last Mile science team, will be co-editors for this special issue. The deadline for submissions is Nov. 30, 2021.
The Amazon Last Mile Routing Research Challenge aims to encourage participants to develop innovative approaches leveraging artificial intelligence, machine & deep learning, computer vision, and other non-conventional methods to produce solutions to route sequencing problems that outperform traditional, optimization-driven operations research methods in terms of solution quality and computational cost.
In this webinar we will:
Define the objective of the challenge
Introduce and explain the data
Explain how participants will be evaluated
Outline the timeline of the competition
Answer questions from participants
Webinar Details
When: March 8, 2021, at 9 a.m. ET Watch the recording hereand download slides.
Participation in this research challenge requires an active full-time affiliation with a university or academic research institute – either as a student or as a researcher/faculty member.
Unfortunately, we cannot accept participants whose full-time occupation is not associated with an academic research institution. This includes consulting, government, and commercial industries to name a few.
What is needed to apply?
Before submitting your application please make sure that you have proof of enrollment available to upload as well as an active academic email address.
Can non-participants access data?
We’re working on making the data and instructions of the challenge available to non-participants after the competition has started. Please sign up for our e-mail list, so that you will get noticed if and when that happens.
How can I form a team with other participants?
Registration for the challenge happens on an individual basis, as we need to validate the eligibility to participate for every participant individually. Approved applicants will get access to a dedicated participant portal shortly before the start of the competition. Within the participant portal, you will be able to form a team with other participants.
