This project aims to improve the ability of 412 Food Rescue to use a volunteer network to redistribute and repurpose perfectly good food destined for the waste system to people in need by adding real-time optimization to parts of their operational processes. Specifically, it will use technology modeled on marked-based systems to optimize both the matching of food opportunities, (e.g., food being disposed of by grocery stores) to distribution-focused nonprofits (e.g., a food pantry or soup kitchen) and the assignment of volunteers to transport the food opportunities. The increased efficiency of these processes will result in more capacity in the system (better use of the transportation resources) while providing a more equitable allocation of food to the network of nonprofits involved in food access.
The mission of the nonprofit organization 412 Food Rescue (https://412foodrescue.org) is to feed the hungry by redirecting food destined for the waste system to people in need. They match food donations to appropriate nonprofit organizations and solicit people through a crowd-sourcing volunteer network to transport the food product. To address the logistically challenges, 412 Food Rescue has developed a base technology platform, called FoodRescueX, that assists in the matching of the food donations and the coordination of its transportation. FoodRescueX, through its phone app and server, significantly improves the visibility of food in the network and the available of people to transport it, while also providing the necessary communication with all participants in order to coordinate the redistribution of the food. We propose to extend and enhance FoodRescueX with automated optimizing capabilities to streamline and improve the decision making in matching food donations to nonprofits and in assigning transportation volunteers to food redistribution opportunities.
The matching of food donations to nonprofit organizations is currently performed exclusively by a human, who uses multiple mediums to announce the availability of particular foods, fields requests over a short period of time, and then selects a matching nonprofit. If there are multiple types of food or large quantities of food, it is largely on the person to determine how best to distribute the food over multiple organizations. For this matching problem, we propose to leverage technology [1, 2, 3] that we have developed over multiple federal programs to create a market-based decision mechanism that uses an automated or mixed-initiative combinatorial auction to determine the best match. The basic idea is that the announcement of available foodstuffs and the ensuing requests (bids) that come back from the nonprofits will be digitized and communicated through the FoodRescueX platform, and then automatically aggregated over a customizable time period. At the end of that time period, the auction is then closed and candidate allocations for the food are automatically generated based on the available food, the requests, and the customizable preferences of the 412 Food Rescue. These candidates are then ranked according to customizable criteria. The selection of the best match can either be done automatically, i.e., based on the highest-ranking candidate, or in concert with a human planner, who is presented the ranked list and their scores from which the person selects the best match. Ideally, over time, the customizable criteria and preferences will be refined and confidence will grow in the automatic selection of the best match, reducing the need for human review.
Once the matching problem is completed, the next issue is assigning a volunteer to transport the food. In current operations, the determination of who will provide the transportation for a food opportunity is a two-step process in which, first, requests are sent to candidate volunteers that are identified based on their availability and proximity to the food opportunity and then, as responses are received from willing volunteers, one is selected typically on a first-come-first-serve basis. This selection process can be improved by considering the existing commitments of volunteers when making the selection. By taking into account their current and future locations over time, including projections of their travel durations, assignment of food transportation tasks will be made more efficient by reducing both the overall travel times of the transporters and the time from when food becomes available to when it is delivered to a nonprofit. Considering existing commitments also helps in identifying candidate volunteers by allowing proximity reasoning to include not only volunteers’ home locations but also where they are slated to be in the future. The resulting more efficient allocations of food transportation tasks to volunteers will expand the capacity of the network, allowing for more opportunities to be served faster. For this optimization, we also propose a market-based mechanism for making these assignments. In this case, the auction starts with when a food transportation opportunity is announced, and then volunteers send back requests (bids) over a customizable time period. At the end of that period, the auction is closed and the bids are ranked according to customizable criteria. For example, if the criteria are to introduce the least amount of travel for volunteers in the network while meeting any deadline constraints, then the ranking procedure would use the expanded notion of proximity that includes consideration of volunteers’ itineraries based on their existing commitments and calculates the distance that would be added to the itinerary of the person if she or he were to service it. As with the matching mechanism proposed above, the ultimate assignment could be done either automatically based on the highest-ranking volunteer or through a mixed-initiative approach that provides the ranked list to a human who then makes the selection. A fully automated decision is likely to be more readily adopted in this context since the assignment criteria is more straightforward.
Our plan is to develop the two proposed market-based decision-making mechanisms within the FoodRescueX technology platform and to run two pilots evaluating each mechanism, one at 6 months and the other at 11 months. If the pilots prove successful, then the expectation is that the enhanced version of FoodRescuesX would go into operation full time.
 Laura Barbulescu, Zachary B. Rubinstein, Stephen F. Smith, and Terry L. Zimmerman. Distributed Coordination of Mobile Agent Teams: The Advantage of Planning Ahead. Proceedings of the Ninth International Conference on Autonomous Agents and Multi-Agent Systems, Toronto, CA, May 2010.
 Zachary B. Rubinstein, Stephen F. Smith, and Laura Barbulescu. Incremental Management of Over- subscribed Vehicle Schedules in Dynamic Dial-A-Ride Problems. Proceedings 26th Annual Conference of the Association for the Advancement of Artificial Intelligence (AAAI 2012), Toronto, CA, July 2012.
 Stephen F. Smith, Zachary Rubinstein, David Shur, and John Chapin. Robust allocation of RF device capacity for distributed spectrum functions. In Autonomous Agents and Multi-Agent Systems 31(3): 469-492 (2017)
7/1/2018 - 8/1/2018 - Refine requirements for both parts of the project, i.e., matching of food donations to nonprofits and assigning volunteers to transport food opportunities, and generate test data.
8/1/2018 - 11/1/2018 - Develop technology matching food donations to nonprofits.
11/1/2018 - 12/1/2019 - Integrate technology matching food donations to nonprofits into technology base at 412 Food Rescue.
12/1/2018 - 1/1/2019 - Conduct pilot of technology matching food donations to nonprofits.
1/1/2019 - 4/1/2019 - Develop technology for assigning volunteers to transport food opportunities.
5/1/2019 - 6/1/2019 - Integrate technology for assigning volunteers to transport food opportunities.
6/1/2019 - 7/01/2019 - Conduct pilot of technology assigning volunteers to transport food opportunities.
Working with 412 Food Rescue, we will first refine the requirements for the project and define APIs between the existing technology base at 412 Food Rescue and the new technology. 412 Food Rescue will then provide test data in the agreed-upon format. We will initially develop each of the two proposed technologies independently, and then integrate each into the existing technology base. After each integration, we will conduct a pilot testing the specific technology in a real-time working context. For each pilot, we will evaluate the results by comparing the options provided by the technology to those of a human planner. We will also measure the number of food opportunities served, with the caveat that, at the current scale of operations, demand may not be pushing capacity such that we would see an increase in that number. This technology will allow 412 Food Rescue to grow to take advantage of more food opportunities without having to depend on having an expansion of their transportation network. At the end of the project, the expectation is that the proposed technology will be deployed into full-time operations at 412 Food Rescue.
Expected Accomplishments and Metrics
The expected accomplishments are 1) more equitable matching of food opportunities to nonprofits based on their preferences, and 2) more efficient allocation of transportation volunteers to food opportunities. Together, these accomplishments will result in better, expanded services in feeding people in need. Both proposed technologies will be evaluated by comparing the decision options provided by the technology to those a human planner would make. Additionally, we track the efficiency of the system in terms of food opportunities serviced, but, at the current scale of 412 Food Rescue's operations, it is not clear that they are oversubscribed and whether or not we will see a bump in these numbers. This technology will have a larger impact as 412 Food Rescue's operations scale up to more food providers and a larger network of nonprofits. If time allows, we will do a coarse simulation of a larger network where we would expect to see the number of food opportunities serviced to increase.
||Carnegie Mellon University
||Faculty - Researcher/Post-Doc
Amount of UTC Funds Awarded
Total Project Budget (from all funding sources)
|Data Management Plan
||June 27, 2018, 3:01 p.m.
||March 30, 2019, 3:34 p.m.
||Sept. 30, 2018, 9 a.m.
||March 30, 2019, 3:34 p.m.
||March 30, 2019, 3:34 p.m.
||May 4, 2020, 11:04 a.m.
No match sources!
|412 Food Rescue
||Deployment Partner Deployment Partner