In San Diego, the defence start-up Shield AI is designing software that will fly drones without any human involvement.
Its software, Hivemind, lets drones “go inside a building without GPS, without communications — these are jammed by adversaries — so it can scope out a building ahead of soldiers”, said Brandon Tseng, Shield AI’s co-founder.
“That’s a huge value added,” he said, explaining that it would open up all sorts of new ways of using drones for law enforcement and the emergency services, as well as the military. Ultimately, he said, he could see his software powering all sorts of commercial drones, such as those carrying out deliveries, for example.
Shield AI is just one of dozens of companies that are working on ways to transform the way drones are used, turning them from piloted point-and-shoot flying cameras into self-flying computers that can work in teams and transmit data straight into the cloud.
While there have been huge leaps in the designs and battery life of drones, the real innovations are coming not in hardware but in their computing capabilities.
A running tally from ModalAI, a unit spun out of chipmaker Qualcomm that sells tech to drone companies, counts about 65 start-ups in the US trying to equip drones with the latest software and artificial intelligence.
Lorenz Meier, chief executive of Auterion, an open-source software platform for drones, said the drone market, which was worth $13.4bn last year, is on the cusp of a revolution.
“It’s a little bit like computing — it was interesting in the ’70s, but they were in some buildings to process tax records,” he said. “But computers changed our lives once they started to talk to each other. And how we use drones will fundamentally change once they start to be networked.”
In June, the Federal Aviation Administration, noting an explosion in the use of registered industrial drones — from 12,000 in 2015 to nearly 500,000 by 2020 — said the industry “appears to be at an inflection point demonstrating powerful stages of growth”.
Meier said he expected drones will shortly operate largely on their own, emerging from a charging station to carry out routine jobs and inspections and flagging relevant data into a Slack channel like any other employee.
“They essentially will become like a person and be part of the conversation,” he said. “So you’ll have real-time footage from a drone in a chat between construction workers. They’ll be like chatbots that start to add information in real time.”
Aerobotics, a South African drone start-up with commercial operations based in Los Angeles, is already helping citrus growers in Florida save their crops from disease.
Its drones fly within a few feet of crops to collect microscopic images that can help farmers determine which ones are growing and which need help.
“We’re using computer vision algorithms to detect that fruit, size that fruit,” said Stuart van der Veen, chief platform officer. “Then, based on production and training data, we project the size of that fruit growing out over a season, and we calibrate that with different collections. And then we calculate the weight estimate, in cartons or kilogrammes.”
He added that by scanning large areas of crops, his drones can learn how to combat citrus greening, a disease, and share best practice. “I think drones are going to save Florida citrus,” he said. “There’s just no other way to collect data like this.”
With all the data Aerobotics has collected, it has even started to write insurance for farmers. “In Florida, we have algorithms which estimate the age of the trees for tree insurance and algorithms to estimate yield for production insurances. [These are] humanly impossible tasks.”
Another innovation has been the advent of “tethered drones” which fly hundreds of feet in the sky but remain connected to a power source for persistent flight.
Easy Aerial, based in Brooklyn, said its tethered “drone in a box solution” can scan miles of forest, national borders or wildfire areas for more than 24 hours at a time. They can be operated automatically or be commanded remotely, anywhere, and even take-off or land on moving vehicles or ships, said Ivan Stamatovski, chief technology officer.
“You can leave in the middle of a desert, push a button and it will go up with a camera and run for a couple of hours, then land,” he said. “Nobody has to be there to man it.”
Among its clients are the Travis Air Force Base and the US Customs and Border Protection agency, which tethers its drone to a giant battery that sits in a truck so the aircraft can run up to 12 hours a day without a generator or external power source.
Easy Aerial has also made a tethered drone it calls the Raptor, which can hover indefinitely. It uses AI to scan from a distance, and upon detection of a person, object, or an incident such as a fire, it untethers itself and flies over for closer inspection.
Mike Ross, product lead at Skydio, a US drone company valued at more than $1bn, said drones were moving quickly along what he called the “arc of autonomy”.
He predicted the next step would be for human operators to fly a fleet of drones simultaneously. “So if I want to go out and build a 3D scan of a building, I can do that with one drone in 30 minutes, or I can do it with two drones in 15, or four drones in less than 10 minutes.”
Adam Bry, Skydio’s chief executive, predicted that within two to four years, cities could be equipped with, for example, a fleet of drones at every fire station that could instantly respond to emergencies. Police forces are already building systems that automatically fly to areas where gunfire is heard and start taking images.
For Shield AI’s Tseng, the shift in focus from hardware to software is likely to give US companies the edge over other countries, such as China, which have so far cornered the market in drone manufacturing.
“[We are at] the tip of the iceberg in terms of capabilities that are going to be unlocked by software,” he said.