Automating Food Drop: The Power of Two Choices for Dynamic and Fair Food Allocation

TL;DR

Automating Food Drop reframes real-time food-donation matching as a dynamic two-sided fair-division problem on a county graph, balancing driver efficiency with fairness to food banks. The authors propose a simple yet powerful two-choice algorithm that assigns each donation to the nearer origin or destination food bank, selecting the one with the least delivered value so far, and prove a $3$-driver-efficient and $(1+\varepsilon)$-multiplicative envy-free guarantee with high probability, leveraging a novel nonuniform, weighted balls-into-bins analysis under $(\alpha,\beta)$-biased sampling. They establish a fundamental lower bound showing trade-offs between driver efficiency and envy-free fairness, and provide tight upper bounds that achieve near-optimal fairness for a broad range of practical distributions and donation weights. The approach is validated through large-scale experiments in Indiana, Indiana++/Midwest, California, and Virginia, demonstrating near-perfect fairness and modest increases in average driver distance, and is deployed as an open-source platform to enable replication and adaptation in other regions. Overall, the work advances both the theory of dynamic fair division with weighted, biased inputs and the practice of scalable, automated food-rescue logistics.

Abstract

Food waste and food insecurity are two closely related pressing global issues. Food rescue organizations worldwide run programs aimed at addressing the two problems. In this paper, we partner with a non-profit organization in the state of Indiana that leads \emph{Food Drop}, a program that is designed to redirect rejected truckloads of food away from landfills and into food banks. The truckload to food bank matching decisions are currently made by an employee of our partner organization. In addition to this being a very time-consuming task, as perhaps expected from human-based matching decisions, the allocations are often skewed: a small percentage of the possible recipients receives the majority of donations. Our goal in this partnership is to completely automate Food Drop. In doing so, we need a matching algorithm for making real-time decisions that strikes a balance between ensuring fairness for the food banks that receive the food and optimizing efficiency for the truck drivers. In this paper, we describe the theoretical guarantees and experiments that dictated our choice of algorithm in the platform we built and deployed for our partner organization. Our work also makes contributions to the literature on load balancing and balls-into-bins games, that might be of independent interest. Specifically, we study the allocation of $m$ weighted balls into $n$ weighted bins, where each ball has two non-uniformly sampled random bin choices, and prove upper bounds, that hold with high probability, on the maximum load of any bin.

Figures (8)

• Figure 1: A typical use case of our platform. 1) A truck driver with a rejected food load wants to go from location A to location B. 2) They fill out a form in the online platform providing relevant details (see screenshot from our platform). 3) The back end runs \ref{['ourAlgorithm']} and matches the request to a food bank. 4) The representative of the matched food bank receives a notification on their phone, via SMS, through which they can access the information regarding the food load (see screenshot from our platform). 5) Once the food bank accepts the delivery, the truck driver is notified and given the contact information of the matched food bank, via SMS.
• Figure 2: An example of a hard instance. Nodes colored red, the set $\{A, v_2, \cdots, v_{n}\}$, are the locations of the food banks. The square nodes ($A$ and $B$) have zero population; all other nodes have a population of $1$.
• Figure 3: Fairness-Efficiency trade-offs for the greedy with cutoff family and \ref{['ourAlgorithm']} in California.
• Figure 4: The first image displays the distribution of the food-insecure population in California. In the subsequent images, the color indicates whether a county receives food near-proportionately (green), more-than-proportionately (brown and black), or under-proportionately (red) to its food-insecure population, under different algorithms.
• Figure 5: Maximum multiplicative envy quickly surpasses $1.5$ and converges to $2$ for the driver optimal algorithm, whereas it monotonically and quickly converges to $1$ for the other algorithms.

Theorems & Definitions (41)

• Theorem 1
• proof
• Theorem 2
• proof
• Theorem 3
• Theorem 4
• Definition 1: $(1+\epsilon)$-choice with $N$ bins
• Definition 2: Majorization
• Definition 3: Stochastic Dominance
• Claim 1