Delivery of goods has always been an important service (online shopping, food deliveries etc.) but our reliance on this service has become even more important during periods of lockdown due to COVID-19.

Have you ever thought about the journey the driver takes when delivering? What is the optimal route between destinations?

A UNSW Canberra PhD student has won third place in a world-wide challenge that attempted to answer this very question.

The Amazon Last Mile Challenge saw more than 220 teams posed the challenge to develop innovative approaches to address the gap between theoretical route planning and real-time route execution [aka the route the driver plans to take and the route they end up taking].

Rasit Abay, part of Team ‘Sky is the Limit’ along with Okan Arslan, Assistant Professor at HEC Montreal, said he was glad to be among the winners in such a tight race with the world's leading mathematicians, computer scientists and route optimisation researchers. 

“What has been found is that there is a gap between periodic route planning and real-time route execution due to road blockages, congestion, parking, etc., so this challenge can foster innovation and research in the last mile delivery now and into the future,” he said.

The aim of last mile delivery is to deliver an item to a recipient in the quickest way possible. Participants in this challenge were provided with 6100 historical route records from five areas across the United States as a baseline and 3000 driver-determined routes. 

The task at the challenge was building models that could identify and predict drivers’ deviations from routes computed with traditional algorithms.

“Our team developed a prescriptive method based on rules that are extracted through descriptive analysis of data and investigated various reinforcement learning approaches to fine-tune the parameters that optimise the metrics that measure the quality of the solution,” Rasit said.

“We wanted to better understand the choices that drivers make on their route, and we measured outcomes by three modelling strategies including effectiveness, interpretability and flexibility.”

By being one of the winners of the challenge, Rasit said that he is thrilled to be one of the contributors of novel research and innovation in last mile delivery for years to come.

“I have been investigating the feasibility of applying machine learning to problems in various fields, including space domain awareness, agriculture/horticulture, remote sensing, and optics.

This is the first time I have had to apply my machine learning (ML) skillset to a combinatorial optimisation problem, which is known to be a particularly challenging field for the application of ML,” he said.

Rasit would like to leverage his experiences in this challenge for space logistics and supply chain and contribute to the booming Australian space ecosystem by bringing his expertise in international data science challenges.

More information about The Last Mile Challenge can be found here.