Using Machine Learning To Feed Families in Need

Community program delivers 6,000 monthly meals with assist from CMU

In this monotonous COVID-19 world, a team of 22 Parkhurst Dining and Eat’n Park employees begin every Monday, Wednesday and Friday the same way. The team arrives at the PNC Firstside Center in downtown Pittsburgh by 6 a.m. and gets to work.

The group makes hundreds of sandwiches, sorts snacks, fruits and vegetables, then packs it all into 400 brown bags. The food isn’t going to restaurant customers or PNC employees. Instead, it is boxed and driven to Penn Hills, a suburb on the east side of Pittsburgh. With help from Carnegie Mellon University, everything will be given away to families in need.

Researchers in Carnegie Mellon’s Robotics Institute used machine learning to create optimal, efficient bus routes that allow community nonprofit organizations to deliver meals to senior citizens, as well as K-12 students and families who would otherwise rely on schools for free meals. The CMU tools identified ideal distribution locations to reach as many people as possible, three days a week.

Nearly 6,000 meals are delivered each month. The program started in July. It has since expanded to also include dinners that feed a family of four, meals that are also produced by Parkhurst at the PNC Firstside location.

Seeded with funding from Mobility21 , CMU’s University Transportation Center, Carnegie Mellon’s Metro21: Smart City Institute leads the project with Allies for Children , United Way of Southwestern Pennsylvania and the Greater Pittsburgh Community Food Bank. The genesis for the idea of route optimization was born nearly a year ago but had to be reimagined because of the pandemic.

"We originally talked to Allies for Children about a plan that would use machine learning to develop cost-effective bus routes transporting charter and private school students across school districts,” said Karen Lightman , Metro21’s executive director. "When COVID closed schools in March and disrupted meal programs around the region, we pivoted. Instead of buses carrying students, we developed a program to have drivers bring lunches to families most in need.”

In Penn Hills, many of the 4,000 students rely on free breakfast and lunch at school. Utilizing a data sharing partnership between the Penn Hills School District and the Allegheny County Department of Human Services, CMU researchers gathered anonymized address information and loaded it into a computer to identify locations and routes.

"The existing bus routes used to transport students weren’t ideal for meal distribution for a number of reasons,” said Stephen Smith , the research professor in the Robotics Institute who developed the delivery algorithms. "Stopping every few blocks isn’t very efficient, and we needed areas where shuttles could safely stop, park and hand out food to groups of people. Our goal was to identify stops and routes to reach as many people as possible.”

Smith’s research focuses generally on core technologies for automated planning, scheduling and optimization. He had previously done algorithmic work to identify efficient routes for snow plows. His research team also produced Surtrac , a novel system for real-time traffic signal optimization that has since been commercialized by Rapid Flow Technologies , a company Smith co-founded. The technology is currently operating in eight North American cities, including Pittsburgh.

Three times a week, the meals prepared at Firstside are transported to the Penn Hills Eat’n Park restaurant, where shuttle drivers from ACCESS Transportation load and disperse them throughout the city from 11 a.m. to 1 p.m. on different routes. They stop on neighborhood streets, in parking lots behind fire stations or at apartment complexes. Signs around the community publicize the stop locations.

Greg Ikper of ACCESS Transportation loads meals before delivering to families.

Sometimes the drivers see new faces. But mostly they see the same people each delivery day. Some are kids at home, learning online. Others are parents and grandparents.

The funds to cover the cost of meals come from the Student and Families Food Relief Fund at the United Way of Southwestern Pennsylvania. That fund has been supported by the United Way, PNC Foundation and the Allegheny County Executive who has allocated CARES Act funding to this work. Funding for ACCESS Transportation is also covered by the United Way of Southwestern Pennsylvania.

Data are collected and reported each day about how many meals are passed out at each location. This allows the organizers to readjust the plans as needed, while giving Smith a chance to tweak the algorithms to improve the process. Organizers are currently working with the Penn Hills School District to adapt the program now that school is back in session, schools are operating under a hybrid model (two days on; three days off) and not everyone is able to return to the classroom.

"Penn Hills is a proud community, which makes it difficult at times to gauge accurately our level of need,” said Nancy Hines, superintendent of the Penn Hills School District. "This project granted access to experts in the field with resources beyond our own, and the strategies that were implemented were done so in a very discreet and dignified manner, which helped us meet family needs which we, in isolation, could never have done. I sincerely thank CMU, Allies for Children, A Second Chance, United Way and our other partners on this very important project.”

A similar program launched in McKeesport, southeast of Pittsburgh, on October 28. Conversations are currently underway to expand the system areas of the region. Carnegie Mellon recently received funding from the National Science Foundation to support this work and expand into other municipalities, to help stimulate future research on these problems and influence remote food delivery problems nationwide.

"This project, helping families in need, is among the most rewarding work Metro21 has ever done,” Lightman said. "It’s wonderful to see the impact on the community, not to mention the letters and feedback we’re getting from families and drivers. CMU’s work is making a difference in people’s lives, which is extremely satisfying, especially during these challenging times.”

Machine Learning Models Identify Kids at Risk of Lead Poisoning


This site uses cookies and analysis tools to improve the usability of the site. More information. |