By Lindsey Choo
Order a package? AI researchers are hunting for the best way to get it to your doorstep.
It might be a drone flying from a nearby pharmacy. It might be a robot that hops off a van that is driving down the street. It might be a system that helps drivers anticipate where a parking spot is about to open up near your door.
In the next few years, AI promises to transform the logistics of consumer deliveries with a number of new technologies and improvements to old ones. Here’s a look at some of the most intriguing innovations on the way.
Putting smarter drones in the air
One of the most expensive and resource-intensive parts of delivery is the last mile—getting packages from warehouses or stores to customers’ homes. Delivery drones promise to make the process more efficient, bypassing the traffic and parking issues that often plague delivery trucks.
The idea isn’t entirely new, of course. Amazon, for one, has been testing drones since 2013, and has delivered packages via drones since 2022. It said last October that it planned on expanding the program this year.
Now a range of companies are using AI to beef up the capabilities of drones. Wing, a delivery-drone company owned by Alphabet, uses AI to let devices decide the best place to leave packages, bypassing obstructions. For example, if a package is supposed to be delivered to a driveway, but the drone spots something blocking the space, it may choose to leave the item on the doorstep instead.
Wing is focusing on last-minute items for nearby customers, like drinks and medication, where the small aircraft can outpace a delivery truck driving out or a customer heading to the store. The average flight time for a Wing drone to customers is under 30 minutes.
Shannon Nash, chief financial officer at Wing, says that the company has seen interest from consumers and U.S. retailers. It recently joined with Walmart to help expand the retail giant’s deliveries in the greater Dallas-Fort Worth area .
The company has plans to enter more retail deals, pending approval to fly in new areas from the Federal Aviation Administration. The drones fly a few hundred feet above the ground, then descend down to about 20 feet to lower packages. The company is also testing a system where drones can do self-assessments, such as battery checks, with little human intervention.
“All roads lead to the autonomous route,” Nash says.
Sending robots to your door
Other companies are tackling the “last 50 feet”—the time-consuming process of getting packages from the delivery truck to the customer’s door. According to research by the University of Washington Urban Freight Lab, this leg of the process accounts for 20% to 50% of overall transportation supply-chain costs.
Vault Robotics, a spinoff from the Princeton University Safe Robotics Lab, is designing robots that can climb curbs and stairs. The robot has a combination of legs and wheels, and a platform with grips on the side to hold a package.
The goal, says Robert Shi, co-founder and chief executive of the company, is to deploy the robots from vans without having to park, eliminating the time spent idle. So, while a van is still moving at a cruising pace, robots can move in and out of the vehicle to deliver parcels to doorsteps.
Predicting parking spaces
Drones and robots are designed to be largely autonomous when making deliveries. But some AI works in tandem with human delivery workers.
For instance, researchers have been looking into the time-wasting process of parking. Giacomo Dalla Chiara, lead researcher at the Urban Freight Lab, says that about 28% of drivers’ time during delivery is used searching for spots.
In a project sponsored by the Energy Department, the lab deployed curb sensors in a Seattle neighborhood, transmitting real-time information about available parking spaces. Combining machine learning and sensor information, the system can predict when the spaces will be available—and direct drivers toward spots that are opening up while drivers are in transit.
Making the drivers safer
Researchers are also looking for ways to use AI to boost driver safety by doing things like alerting drivers to dangerous situations much more quickly and accurately than traditional onboard systems.
In a study with the Virginia Tech Transportation Institute, a dashcam from fleet-management tech company Motive successfully alerted drivers to unsafe behavior about 86% of the time. The devices looks for red flags such as drowsiness by monitoring drivers’ facial expressions and motions.
Motive also offers AI technology that is designed to eliminate blind spots, says Abhishek Gupta, vice president of product at Motive. Using the AI omnicam, cargo drivers can see on the sides and rear of their trucks through the camera, and the AI can look for signs of potential crashes and alert drivers.
Optimizing routes in real time
Delivery drivers already use software to find the best routes. Researchers are looking to beef up that capability with AI for that job.
Matthias Winkenbach, director of the MIT Megacity Logistics Lab, is working on a model that can take into consideration complex real-world constraints. For example, drivers can choose a route that may not be the shortest but allows them to park more conveniently or unload packages in safer spaces.
Vault Robotics, meanwhile, is trying to help its robots figure out the best—and safest—routes from delivery vans to customers’ doorsteps, says Jaime Fernández Fisac, co-founder of the company and principal investigator at the Safe Robotics Lab.
For example, instead of taking a straight path to a home from the van, robots can learn to avoid bumping into humans walking or an animal running, and take an alternative route to avoid mishaps. Fisac says they are training the robots’ decision-making with millions of situations that will then pass through real-time testing and verification.
“It’s like when Doctor Strange is about to fight Thanos,” Vault Robotics’ Shi says, referring to how AI can make sure robots take routes that are efficient but safe. “He’s on his little rock, looking at 14 million different potential futures, planning for all the potential cases that can happen, and then choosing the best outcome.”