This article was published on May 16, 2022

Exclusive: John Deere closes in on fully autonomous farming with its latest AI acquisition

Some driverless vehicles work harder than others


Exclusive: John Deere closes in on fully autonomous farming with its latest AI acquisition

John Deere is announcing the acquisition of a state-of-the-art algorithm package from artificial intelligence startup Light.

For those of you wondering when driverless vehicles will truly begin to make their mark on society, the answer is: today.

Up front: No, you won’t be seeing green tractors rolling themselves down city streets anytime soon. But the timeline for fully autonomous farming is being massively accelerated. Today’s purchase is all about John Deere’s need for speed — and accuracy, but first let’s talk about rapid development.

I spoke with Jorge Heraud, John Deere’s VP of Automation and Autonomy, and Willy Pell, VP Autonomy and New Ventures at Blue River Technology (a John Deere company).

The <3 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

They explained that this acquisition will not only accelerate the development and deployment of the company’s AI tech, but it’ll also allow the equipment to literally move faster, safely, without human intervention.

Background: Light, the company Deere’s partnering with for the asset purchase, is a major player in the autonomous vehicle field. It uses a computer-vision approach to self-driving that allows the AI system controlling a vehicle to ‘see’ the world similarly to the way biological systems do.

In essence, Light’s algorithms will allow Deere’s equipment to use industry-standard cameras (read: regular old off-the-shelf vision systems) to achieve virtually unparalleled depth perception.

This is similar to the approach that Tesla takes for its ‘Full Self-Driving’ (FSD) system.

We’ve criticized vision-only approaches in the past here at Neural, but this is different. Passenger vehicles have to travel in busy thoroughfares where the slightest mistake can result in the loss of human life.

On a farm, the stakes are much different. Tractors and other agricultural equipment need to be able to identify crops and obstructions in order to ultimately achieve the goal of optimizing food output.

The current industry-standard solution for autonomy (usually) involves using a combination of LiDAR and computer vision. This allows developers to achieve the resolution and depth necessary to, for example, teach an AI system to ‘see’ a pedestrian walking across the street in a snowstorm.

But that’s not necessarily the best way to go about things in the agricultural space. Deere’s vehicles, for example, often need to be able to see individual weeds in real-time as the vehicle is moving across bumpy terrain.

A little deeper: LiDAR is costly and doesn’t really allow for the close-up fidelity necessary for agricultural operations. It’s nice to be able to see people walking across a street from hundreds of yards away, but it’s not very helpful for planting individual seeds, killing individual weeds, or alerting farmers to problems specific to their fields.

Deere’s autonomous vehicles have to be more like robot workers than mere transportation vessels. And, for that, they need to focus on sensing at speeds and resolutions that suit client needs.

As Pell told me:

The perfect sensor is a camera that gives you LiDAR-quality depth.

Unfortunately, cameras typically need to be motionless in order to properly process light for depth. Modern systems overcome this through the use of image-stabilization algorithms.

But it’s one thing to figure out how to compensate for your shaky grip when you’re posing for a selfie or teach an AI model to recognize stop signs.

It’s an altogether different problem to keep a camera pointed in the right direction as its steel mount twists and vibrates from the force of thousands of kilograms of machinery bouncing around on uneven terrain.

And that’s just the tip of the iceberg. Light’s technology will allow Deere to compensate for all these problems using industry-standard cameras. That means the company can keep hardware costs low by applying cutting-edge algorithms to gear it potentially already has.

Going forward: Heraud told Neural that the company expects the acquisition to start paying dividends for Deere customers within the next few months. He says the company eventually intends to achieve higher speeds with its current autonomy systems and to include more vehicles.

The ultimate goal is to fully-automate and optimize farm work so that farmers can spend their time on higher-level management and other uniquely-human endeavors.

Quick take: Deere’s already made quite a splash in autonomy and automation. Less than a month ago we said it was slowly becoming one of the most important AI companies on the planet. But this acquisition gives us every reason to update that assessment — it’s now quickly becoming the AI company to watch.

Everyone needs to eat. And, despite the general public’s perception that AI will put people out of work, there’s an ongoing labor shortage in the agricultural world — machines can help.

It’s hard to think of a better use-case for autonomy than optimizing humanity’s ability to feed itself. And, just as exciting, it’s also hard to think of a better and safer testing ground for automation and autonomy than the wide-open spaces of a giant agricultural plot.

If nothing else, this acquisition signals John Deer’s further emergence as an artificial intelligence company at the forefront of the ag-tech industry.

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Also tagged with