Skip to content

Regina company's drone harnesses artificial intelligence

The flying sprayer has the ability to identify and treat individual weeds as it makes its way over a broad acre canopy.

LANGHAM — An autonomous, fixed-wing drone that uses artificial intelligence to identify and treat individual weeds within a broad acre crop canopy was on display at the recent Ag In Motion farm show near Saskatoon.

This was the first time Regina-based Precision AI displayed its new platform for the public, although it had shown earlier prototypes including a smaller drone and a ground-based robot at previous farm shows in the province.

The new drone has a 20-foot wingspan, weighs 55 kilograms empty, and has a 20-litre application tank.

The Precision AI spray drone was designed to treat individual weeds while travelling 80 kilometres per hour at a height of eight to 10 feet.

Precision AI designed the drone to take off vertically like a quad-copter drone, then the back propellers shift 90 degrees to push it in a forward direction like a plane while the front propellers turn off.

Daniel McCann, chief executive officer of Precision AI, said developing technology capable of detecting and then hitting individual weeds with herbicides from a fixed-wing drone is not for the faint of heart.

“It’s insanely difficult. It’s probably about a 9.5 out of 10 on the difficulty scale,” McCann said at the Precision AI booth at the Langham, Sask., event held July 19-21.

“We’re building an electric plane and then on top of that we have to build the AI models that can recognize the problem. We even get them to operate roughly 10-times faster than industry standard and that alone is a challenge. And then, once you get all that working together, then you actually have to be able to trigger the spray to actually precisely hit the plant while you’re moving.”

Over the past few years, multiple companies have developed technology capable of identifying and treating weeds within a crop canopy, which is known as green on green, as well as green on brown products that can identify and treat weeds in a fallow or a pre-plant scenario.

Developing this technology has been challenging because it has to offer a better return on investment (ROI) compared to blanket-herbicide applications.

I’ve talked to a few technicians who are building green on green system for ground-based sprayers, and they said selective spraying systems need a hit rate of the in-crop weeds to be above 90 percent to achieve a better ROI compared to blanket applications.

This threshold isn’t static because it matters which crops and weeds we’re talking about.

However, adding the increased difficulty associated with aerial application, including higher application heights, speed and turbulence caused by the drone, will make it even harder to maintain an acceptable hit rate of the in-crop weeds.

It will be cheaper to fly the Precision AI drone over a field compared to covering the same field with a ground sprayer, so it may be the case that it won’t require as high of a hit rate to provide a competitive ROI.

The most challenging aspect for a fixed wing drone capable of green-on-green applications is the development of the AI capable of identifying individual weeds at such high speeds.

Precision AI uses advanced processors by Nvidia Corp., which used its expertise in building graphics processing units (GPUs) for the graphics and gaming industry to lead the development of the AI chip market.

The processing power required for the Precision AI drone to conduct green on green spraying is available, and the company has spent tens of millions of dollars developing AI-based programs to perform this task.

“We’re already ahead of the game in a lot of ways. We’ve invested very heavily into what we believe to be the best AI in the industry, both in terms of speed and performance,” McCann said.

“A lot of the competitors out there are focused on only corn and soy, as an example. We support eight crops, including some of the bigger crops grown in Canada.”

Precision AI is training its AI to perform green on green spraying in corn, soy, canola, lentils, peas, barley, wheat and oats.

Another challenge in building the Precision AI spray drone was sourcing cameras capable of producing sharp images at high travel speeds.

“We naively assumed that we could use off the shelf cameras. No, not at the speeds we’re trying to go. There’s a trade-off between the speed you go and the quality images because the faster you go, you’re either going to get motion blur or if you have really short exposures… you don’t get enough light in the camera,” McCann said.

“We had to develop our own custom platform for the actual cameras themselves and then make sure it can go fast enough through our artificial intelligence system to be able to recognize the plants in that image. So, it was a really big challenge.”

McCann said Precision AI decided to develop a fixed-wing drone instead of a quad-copter style platform because it is much easier to control where the spray will land.

“This has the same type of drift profile as an aerial applicator so we can actually work around those rules by modelling this after a well-known aerial applicator problem,” McCann said.

A “key problem is the drift when its (copter-style drone) propeller wash that points downwards to keep the drone in the air that also blows your spray in unpredictable ways. This is actually one of the leading reasons why drone spraying is prohibited in a lot of regions because nobody really understands the drift profile,” McCann said.

There is a big difference between the drift profile caused by different styles of drones, however when applying fungicides, the downwash caused by the propellers may help penetrate the crop canopy.

Fixed wing drones also have better flight time and payloads compared to copter style drones.

“You can fly for long distances because the wing lift generates enough lift to carry that spray without draining your batteries,” McCann said.

Battery capacity is a limiting factor with most agricultural drones, so to get around this problem a gas-powered generator was installed to power the electric propellers on the Precision AI drone.

“It can stay in the air for two and a half hours; it’s designed for broad-acre. If you have to change your batteries every five to 10 minutes than you can’t do broad-acre,” McCann said.

To help make this spray drone feasible for broad-acre crops, Precision AI is also developing a hive system to store and transport up to four drones, which can also automatically refill and refuel the drones for continuous deployment.

Operators of the drone will input the job they want it to perform, then the drone will complete the task autonomously once it’s given the command to begin.

Adherence to Transport Canada rules for drones, including the need to maintain a line of sight to the drone by the operator, are required to fly the Precision AI drone.

There are no crop protection products typically used for broad-acre production in Canada that have drone application on their label, but if there were, operators of this drone would need an aerial applicator’s licence before they could apply it.

A working group has been established in Canada to help update and establish rules for drone applications of crop protection products, which included participants from the Pest Management Regulatory Agency, Transport Canada, the academic community, agrochemical companies and representatives from the drone industry.

The Precision Ai spray drone is undergoing field trials in Ontario this summer.

During Ag In Motion, Precision AI took $100 pre-orders for its drone, and McCann said they will likely have production units available for the 2024 growing season.