We recently invested in Tensorfield Agriculture, an early stage robotics company that’s moving us closer to a world free of herbicides.
The company fits nicely in our investment thesis for a number of reasons. First, the founders are PEAs—they’re passionate about solving a specific problem and have the proven chops to pull it off. Second, they’re targeting the agriculture industry, which, as we’ve written here before, is one that’s ripe for data disruption. Third, they’re following a RaaS business model.
The four co-founders—Xiong Chang, Sandeep Mirchandani, Louise Thomas and Cheehan Weereratne—bring an impressive array of talents and expertise in robotics, integrated circuit design and computer vision with deep learning. They’re highly specialized computer scientists and engineers who trained at Stanford, Cambridge, Imperial College (where they met) and Simon Fraser University.
They also have deep ties to the manufacturing ecosystem in Shenzhen, China, having taken part in the same SOSV/HAX Accelerator as ViaBot co-founder and CEO Gregg Ratanaphanyarat (who introduced us).
While in college together, they got passionate about the problem of weed-killing at industrial farms.
Farms often have two options for controlling weeds that both come with significant challenges: They hire people to pull them (which is getting more expensive) or they spray harmful herbicides designed to kill weeds and not crops (at a significant environmental cost).
Several university studies—from UC Davis and the University of Bonn-Landtechnik in Germany—validated a new technique using heated canola oil that produces a better outcome for the farmer. The heat energy penetrates the meristematic cells of young weed plants, killing them more effectively than chemical-based insecticides without the negative environmental impact.
The team developed a robot that can roam rows of crops, accurately identify weeds growing near the crops, and precisely deliver a shot of hot oil, killing only the weeds. The key to making this possible is the ability using vision and AI to very accurately identify weeds in real time and target them accurately. (Click here to see a video of the robot in action.)
When we were first introduced to the team, we were impressed by their early prototype and their scrappy approach to building it. Like Ratanaphanyarat, they turned to Craigslist for simple parts (in this case, using shock absorbers from old mountain bikes) and kept things simple.
We were also impressed that they’d amassed a lot of market data from talking directly to farmers.
The business opportunity was obvious from the start: Farms largely rely on manual labor to do much of their weed picking—and that labor is getting much more expensive and harder to find.
An executive of one the Bay Area’s largest farms reported to us that the company spends more than $6 million on weed picking alone and that Amazon’s $15 an hour warehouse jobs, immigration shifts and post-pandemic trends are sapping farming labor supply.
Robots as a service
Of course, the company has competition in the space. The RaaS model gives Tensorfield a leg up, though. The team doesn’t have to sell an expensive piece of equipment to a farmer who then must maintain it and hope that it doesn’t become obsolete in a few years. Rather, they rent out the machine on a monthly basis and rotate it from farm to farm.
As we wrote in “The Rise of the Autonomous Machines (Part 2),” we look for companies that can return the cost of their robot in less than a year of leasing it out. Tensorfield falls well within that window, which will allow the company to continue to develop and improve on its technology using monthly revenues.
From there, Tensorfield is in a great position to add new services like crop monitoring that farmers already are asking for.