The Center for Data Innovation spoke with Steven Snyder, co-founder and director of robotics at Stout AgTech, a California-based startup that uses AI to help farmers automate high-intensity field work, such as weeding. Snyder discussed the inspiration behind Stout, the challenge of collecting good data to train agriculture-specific AI models, and how differences in farming regulations across the world drive innovation at Stout.
Martin Makaryan: What motivated you to help found Stout AgTech?
Steven Snyder: Stout works in the agricultural technology space and was born from a realization that there is a significant labor shortage in the field. My partners and I had a background in working with large farms and shipping companies, and we had seen how labor shortages in farms affected their productivity. Labor scarcity in agriculture is a problem across the world, which makes the need for innovation in this field even more important. After recognizing this fundamental problem, we thought that an AI vision system that could automate some of the most labor-intensive tasks that growers need to complete would solve the problem. This was the vision that pushed me to co-found Stout.
Since its founding, Stout has been able to address a key challenge that traditional computer vision products in this space alone do not solve—farmers still have to manually address the problem crops these systems identify. With significant AI advancements over the past few years and a tedious process to collect the necessary data, we were able to automate the process for a variety of labor-intensive tasks like removing weeds from the field or fertilizing the soil. AI has been a game changer as our tool easily distinguishes between weeds and different varieties of crops, allowing growers to not only increase their productivity through automation, but also collect information about their crops that can help them make data-driven decisions.
Makaryan: Can you explain the technology that powers Stout’s smart farming tools?
Snyder: Our smart cultivator is a tool attached to a tractor that can visually detect weeds and crops using AI-enabled software. The tool automatically removes weeds using attached mechanical blades and collects data while operating. In terms of power, the smart cultivator uses electricity through a hydraulic motor that we also attach to the tractor, but it comes with a battery that can last a few hours in case of emergency.
On the user side, there is minimal involvement to run the cultivator. Once the user turns it on, they just need to enter some basic information, such as the crop type, how far the seeds are apart from each other, and the horizontal distance between the seed lines. Multiple cameras under the machine take images of crops and the cultivator uses the user information to detect and remove the right weeds. An iPad screen inside the tractor allows the user to always monitor the process and change the settings if they wish to. Aside from adjusting how close the blades should get to the crop to avoid damaging the roots, the user does not need to do anything else to operate the cultivator. Our algorithms do the rest.
I think one of the biggest things that sets us apart is the fact that we make both the hardware and software powering our smart cultivator. This has allowed us to improve the technology as we have expanded into new regions and worked with more clients. The expertise to build the hardware is not just a convenience for us or our customers, but an important feature of our business model. Because we are not contracting out, we have the in-house expertise to perform integrations that carefully balance between hardware and software. For example, when it comes to weeding blades, we can adequately assess when we can fix an issue by writing code that changes how the tools function or when we actually need to make a change to the hardware. The fact that we have control on the nuts and bolts of both allows us to make decisions that address the specific needs and problems of our clients.
Makaryan: How did you overcome the challenge of collecting data when initially training your AI models?
Snyder: People get very excited about the utility of AI products without sometimes realizing the amount of work that goes into collecting the data that makes these products possible. In our case, the challenge was even more complex because our task was not to simply collect thousands of images of crops to train our models. There are a number of nuances that are specific to agriculture. For example, new varieties of crops come out every year, or weeds move into new regions. These are changes that we need to consider when making models that can accurately detect and make decisions on what to remove.
When we started, there was not enough data to use. The datasets that existed were not high quality, which is why we decided to create our datasets from scratch. We started with a cohort of interns who would collect images from 10 to 15 fields a day, while our labeling team, which had to build the expertise to correctly identify hundreds of different crops, labeled them and entered them into a dataset. This was a very intensive effort, and I am proud that today, we have over 10 million examples of crops and various plants in our dataset. I think this may be the largest agricultural image dataset in the world.
Makaryan: Do any regional or country-specific differences impact your technology?
Snyder: The difference we see usually comes from the variety of farming regulations that growers need to contend with. Sometimes, specific regulations can also impact the needs of farmers when it comes to smart technologies such as ours. For example, the European Union has different standards than the United States regarding the use of fertilizers and pesticides. Because European regulations place limits on how much fertilizer farmers can use, growers in these regions would benefit from using precision dispensing equipment. That is why we have added a smart rate fertilizer to our cultivator that uses the same AI vision system to determine how much fertilizer to apply to each crop. This addition reduces the cost of fertilizers for farmers, ensures compliance with regulations, and benefits the environment by minimizing the amount of chemicals that growers release.
Makaryan: How do you think future breakthroughs in AI will impact Stout AgTech?
Snyder: We are heavily dependent on AI technology. Whether writing a new code, developing our engineering tools, or enhancing data analytics, any improvements in AI will definitely benefit us. I think the main improvement from new breakthroughs would be in data analytics and providing insights to our clients regarding the performance of our smart tools. Of course, depending on what kind of breakthroughs we will see, we can also enhance our product usability by improving visual detection or streamlining automation.