A team of researchers announced on Mar. 12 the development of an artificial intelligence-based model designed to improve the fairness and efficiency of humanitarian aid delivery during natural disasters. The approach uses a combination of trucks and drones to overcome damaged or inaccessible roads, with trucks transporting large quantities of supplies as close as possible and drones handling last-mile deliveries.
The research addresses a key challenge in disaster response: ensuring that aid reaches all affected individuals quickly, including those in isolated areas. Stevens Associate Professor Jose Ramirez-Marquez said, "This ensures that aid reaches even the most isolated and inaccessible locations." He explained that drones can be repeatedly resupplied by trucks, allowing for continuous delivery to hard-to-reach places. "This collaborative approach combines the strengths of both vehicles, where trucks handle the bulk transportation of goods while drones extend the reach to remote or difficult-to-access locations," he said.
Ramirez-Marquez sought to further improve this process by developing a mathematical model focused on minimizing the time it takes for the last person to receive assistance. Rather than just reducing average delivery times, his goal was to make aid distribution more equitable. "That way, aid is spread more evenly over time," Ramirez-Marquez said. "And it's also a fairer way to deliver aid."
Working with Stevens Teaching Assistant Professor Nafiseh Ghorbani-Renani and PhD candidate Ramin Talebi Khameneh Ramin, Ramirez-Marquez used AI tools such as evolutionary algorithms to optimize service fairness, workload balance, and operational costs. "We used the so-called evolutionary algorithm, because it evolves from one generation to the next," Ramirez-Marquez said. "With each iteration it tells us, 'oh, I found this other solution, and I found this better solution.' At the end, we look at all the good solutions and say, 'you know, this is the best solution we found.'"
The team tested their system using simulations based on real-world disasters: severe flooding in Hoboken, New Jersey during Hurricane Sandy in 2012 and flash floods in Hopkins County, Kentucky in 2025. They also examined how potential disinformation could affect prioritization strategies for equitable access after disasters. "This framework does not infer or detect disinformation," Ramirez-Marquez clarified. "Instead, it evaluates how prioritization strategies can safeguard equitable access when information may not always be correct in the disaster aftermath."
Their findings were published in March 2026 in Computers & Industrial Engineering under the title Multi-objective optimization of a truck–drone delivery system for fair and efficient humanitarian logistics under disruption and disinformation.
Looking ahead, Ramirez-Marquez said emergency responders could use this algorithm during future disasters by generating optimal routes as situations evolve and needs change. He added that testing with municipalities would be an important next step: "The algorithm is ready to be used, now we just need to test in real world settings."
