Mineral: Applying Silicon Valley ‘Superpowers’ To Agriculture
For the past five years, the team at Mineral have been working inside X, the moonshot factory of Alphabet (Google’s parent company).
Today, Mineral graduates as the business is made a standalone company within Alphabet.
What’s the first thing to come from Mineral to the farm?
The company has been working through established ag partners (more than a dozen currently) to develop tools with machine learning, artificial intelligence, robotics and geospatial technologies.
“We are hoping to bring a new perspective as an outsider by partnering with those who have deep expertise as existing players in the industry,” says Mineral CEO Elliott Grant. “If we are successful, we will help build a fundamental step change in capabilities with a set of tools that are incredibly powerful.”
Grant shares Mineral wants to apply its expertise a large scale, while respecting the details and anomalies inherent in agriculture. Farmers won’t buy technologies direct from Mineral, but the company’s goal is to have its tools embedded in the inputs, equipment and decision-making tools used by farmers.
“Every farm looks a little different, and every plant looks a little different. We know the models we build have to work across 100,000 different environments, but our strength is in steadily investing in the diversity of data. Technology can handle all these situations when the model built are robust and are built to handle the complexity and analyze the diversity of data in agriculture. It’s our job to bring the user along the ride.”
He says the first product a Mineral partner is likely to bring to market will be around weed detection tools. The team at Mineral says their work spans scouting, identification, mapping, and application.
“One of the areas we're already seeing a lot of demand for our partners is our ability to identify and model weeds,” says Greg Chiocco, Mineral growth lead. “It’s an area we spend a lot of time and energy on as the implications of weed modeling are endless.”
The data Mineral is collecting is another of its core competencies.
With its work, Mineral has developed an image database of more than 17 crops in every stage of growth in multiple environments. It’s being used for modeling and can be used in many projects with partners.
“Much like an Android model, we aren’t one fully closed ecosystem. We have an open ecosystem,” says Mineral Chief Operating Officer Erica Bliss. “We think about how ag is evolving, and it’s not about just one set of tools. We think it’ll take this open approach to tackle to the nuanced and local challenges.”
A visual example of the work Mineral is doing is the rover platform.
The rover is a four-wheeled semi-autonomous platform Mineral has developed over four years. There are multiple configurations of the machine from narrow row configurations to high clearance. The core functionality of the rover is as a data collection machine.
“Think of this as our Google Streetview car,” Grant says. “It’s not a final product. It’s a research tool.”
The rover can collect 1 million image a day to measure characteristics of plants. For example, in one 24-hour period one rover collected nearly 6 million images measuring a dozen plant traits for R&D purposes.
One company Mineral has said its partnering with in the input space is Syngenta.
Another partner in the specialty crop business has yielded multiple projects moving forward.
Driscoll’s has been a partner of Mineral to help it solve two problems: in-field harvest analysis for its berry crops and post-harvest crop condition ratings.
While harvesting strawberries and raspberries comes with challenges on assessing crop condition, a specific technology has been developed to monitor the crop in the field.
“It wasn’t just that we increased the yield harvested, the human experts got better,” says. “They got better insights on yield forecasting so they learned why the machine was making the prediction.”
Grant adds, “It may surprise some people that artificial intelligence isn’t replacing people. Rather, it’s a co-pilot to help decision making by pointing out anomalies.”
The second project illustrates how artificial intelligence is used to assess the conditions of harvested berries including color, size, bruising and more.
This application can be run on a mobile phone. It only requires a blue background and a color key card for calibration.
One focus for the team at Mineral is to make its technology easier to use in the field.
With its partners at Alphabet providing proprietary chips, Mineral is excited to accelerate the use of Edge Computing—meaning there doesn’t have to be a strong, consistent internet connection to run computer applications, collect images, make maps and make decisions.
Grant shares how the devices have gotten smaller and smaller in the past five years.
“Something that can be held in two hands is able to provide stereo imagery, its own lighting, computing power, GPS and power management for the rover,” he says. “It’s getting lighter, cheaper and more powerful every year.”
In fact, they are working to build products for situations without connectivity.
“We shouldn’t assume high bandwidth is available,” he says. “Doing this work on The Edge for the next decade is critical.”
What about other well-known Google technologies?
When asked about geospatial, Grant says the team at Mineral benefits from the work done at its sister company Google Earth. He adds Google Earth has been an amazing tool for researchers and has moved the needle for the GIS system.
“We benefit from generating data from the Google Earth engine and built a layer translated to ag specific with field boundaries, crop types cover crops and tillage,” Grant says.
Mineral has analyzed 10% of the total farmland—which means of the 4 billion acres on Earth for farmland. 450 million acres of farmland are in the Mineral platform.
“Scale is in our DNA, and one our superpowers is scale,” Grant says.
While Google Glass has so far been taken off the shelf for the consumer market for more industrial applications, Grant doesn’t rule it out as a future platform for innovation.
“I think there’s a future in which the interaction isn’t intermediated by a phone. Perhaps it’s headworn. The computing power is there. But if it will be a face worn AR device, I don’t know. But the fundamental tech is there,” he says.
So what's Mineral's vision for the future of agriculture?
Grant says, “Five years from now, I would hope the tools we imagine, a machine learning enabled co-pilot for example, is no more remarkable than using Google Maps for navigation.”