So What is a Crop Robot?

We at Better Food Ventures and The Mixing Bowl create various market landscape maps that capture the use of technology in our food system. Our intent in producing these landscapes is to not only represent where a technology’s adoption is today, but, more importantly, where it is heading. So, as we developed this 2022 Crop Robotics Landscape, our frame of reference was to look beyond mechanization and defined automation to more autonomous crop robotics. This focus on “robotics” perhaps created the hardest challenge for us—defining a “Crop Robot”.

According to the definition of the Oxford English Dictionary, “A robot is a machine—especially one programmable by a computer—capable of carrying out a complex series of actions automatically.” Putting agriculture aside for a moment, that definition means that a dishwasher, washing machine, or a thermostat controlling an air conditioner could all be considered robots, not things that evoke “robot” to most people. When asking “What is a Crop Robot” in our interviews for this analysis, the theme of “labor savings” came through strongly. Must a crop robot be a labor reducing tool? This is where our definition of a crop robot started us down the “It depends” path?

• If a machine is only sensing or gathering data, is it saving labor enough to be considering a robot?

• If a machine does not have a fully autonomous mobility system to move around—perhaps just an implement pulled by a standard tractor—is it a robot?

• If a machine is solely an autonomous mobility system not designed for any specific labor-saving agriculture task, is it a robot?

• If the machine is an unmanned aerial vehicle (UAV)/aerial drone, is it a robot? Does the answer change if there are a fleet of drones coordinating amongst themselves the spraying of a field?

Eventually, for the purposes of this robotic landscape analysis, we focused on machines that use hardware and software to perceive surroundings, analyze data and take real-time action on information related to an agricultural crop-related function without human intervention. 

This definition focuses on characteristics that enable autonomous, not deterministic, actions. In many instances repetitive or constrained automation can get a task completed in an efficient and cost effective manner. Much of the existing and indispensable agricultural machinery and automation used on farms today would fit that description. However, we wanted to look specifically at robotic technologies that can take more unplanned, appropriate and timely action in the dynamic, unpredictable, and unstructured environments that exist in agricultural production. That translates to more precision, more dexterity and more autonomy.

Traversing the Valley of Death to Achieve Scale

The Government of the United Kingdom recently released a report that reviews Automation in Horticulture. In the report they include the automation lifecycle analysis graphic shown below that they refer to as “Technology Readiness Levels in Horticulture”. If we were to map the more than 500 companies we researched in our analysis, well over 90 percent of these companies would still be labeled in the “Research” or “System Development” phases. Historically, many agriculture robotics companies have failed to succeed, perishing in the “Valley of Death”. Only a handful of companies have reached “Commercialization”, a phase where companies attempt to traverse the perilous journey from product success to business success and profitability.

There are many reasons why ag robotics has had a high failure rate in reaching commercial scale. At its core, it has been very difficult to provide a reliable machine capable of providing value to a farmer on par with a non-robotic or manual solutions at a cost effective price point.

Amongst the technical challenges crop robotics face are:

1. Design: In the early days a company may want to vary its product design to try new things. But at some point as it begins to scale, it needs to lock in standardization to the degree possible. Updating deployed systems remains a continuous challenge. 

2. Manufacturing: Maturing companies move from custom to standardized manufacturing. One company we spoke with had gone from building machines itself, to just building a base and then having vendors doing sub-assembly. Now they have gotten to a point of maturation that not a single team member touches a wrench as all manufacturing is done by partners. 

3. Reliability: A metric usually used is hours of uninterrupted operation and scaling requires going from “faults per mile” to “miles per fault”. The ability to handle the adverse and unpredictable conditions of agricultural production exacerbates the difficulty in creating a reliable machine. As an example, one person told about the unforeseen challenge of working in vineyards where the acid from grape juice accelerates equipment deterioration. 

4. Operation: At some point in the scaling process, farm staff will operate the machine without the presence of robotic solution provider support staff. At this point, there are often knowledge gaps on how to effectively operate the machine that need to be resolved. A step in scaling is getting farm staff trained to operate the machines themselves.

5. Service: Another metric we heard was about decreasing service support resource requirements: How could a robotics company switch from having X number of people support a single unit to having a single person support Y number of different units? 

A last technical facet of scaling is the ease with which a platform can be modified to serve multiple crops or multiple tasks. The space is still so early that we don’t have that many data points about repurposing technology for multiple crops/tasks. However, it is something many companies are obviously looking to prove to upsell customers or convince investors they have the potential to serve a larger market.

We heard from numerous crop robotic startups and investors that the technology challenges need to be tackled first, then the economic and business challenges can be addressed. The reality, of course, is that a successful crop robotic solution developer must face several challenges simultaneously: sustaining a business while refining product-market fit to get paying customers; refining product-market fit while sustaining the interest of investors; and sustaining the engagement of farmer customers.

On the business side, we tried to identify when a company could claim it had made it through the “Valley of Death”. One group we spoke with very simply said there were three key business questions to ask:

1. Can we sell it?

2. Does demand outstrip supply?

3. Do the unit economics work out for all parties?

To the question of “Can we sell it?” The answer usually equated to when and if the robot could perform the task on par with a human—a comparable performance for a comparable cost. That performance clearly varies by crop and task. As an example, there was a generally shared sense that “picking” was the most difficult task to achieve on par with the time, accuracy and cost of a human.

One thread that came up in our conversations is that many farmers may not yet see the longer-term potential of what robots can do in agriculture. They look at (and value) them merely as a way to replace the tasks a human does—but do not look at what more efficient approaches beyond the capabilities of humans that could be enabled with these powerful platforms.

In our discussions we probed on whether the business model of a crop robotics company made a substantial difference in whether they could sell it. Responses were wide-ranging as to whether there is a benefit to having a “Robotics as a Service” (RaaS) model versus a machine buy/lease model. Our net conclusion regarding business models is that, while it may be advantageous to offer “Robotics-as-a-Service” (RaaS) in the early stages of a company’s development, over the longer run companies should plan to operate under both a buy/lease and a RaaS model. The advantages of RaaS in the early days are that they 1) allow a farmer to “try before you buy” which lowers the complexity and cost, and, thus, lowers the barrier to adoption and 2) offer a startup to work more closely with farmers to understand problems and identify potential new challenges to solve.

Many startups have “hyped” their solutions too early, before they could conquer the many complexities involved with successfully operating in the market. This “hype” has caused many farmers to be leery of crop robotics in general. Farmers just want (and need) things to work and many may have been burned in the past by adopting technologies that were not fully mature. (As one startup said, “It is hard to get them to understand the iterative process”). Still, farmers are also known as problem solvers and many do engage with startups to help mature solutions. 

Of course, the “Can we sell it?” question should really be extended to “Can we sell and support it?”. An interesting point to watch between incumbents and new solution providers will be the scaling of startups and the resulting need for those companies to have a cost-effective sales and service channel. Incumbent vendors, of course, have those channels, and John Deere and GUSS Automation have announced just such a partnership.

Like farmers, investors also walk hand-in-hand with a robotics startup crossing the Valley of Death. Investor sentiment toward agriculture robotics is mixed. On the one hand, there is an acknowledgement that there have not been notable exits of profitable startups in this space (as opposed to those just having desirable technology). On the other hand, there is a recognition that agriculture’s labor issues are becoming more acute and large potential markets could be realized this time around. Investors also see that the quality of the technology and startup teams have improved in the last few years.

It is encouraging to see more investors looking at the space than a few years ago, writing bigger checks in later rounds, and investing at high valuations. Investors also understand the challenges better than before so that they can differentiate between segments developers are targeting, e.g., the difficulty of harvesting in an open field versus a greenhouse.