Indoor farming is a mature and highly effective form of food production that is much more widespread than the common consumer might realize. Despite the recent interest in organic foods and local produce, most Swedes do not realize that many staple foods we pick up at grocery stores are e grown almost exclusively indoors! This transformation has been driven primarily by technological advancements in irrigation, robotic automation, and other physical facets of farming.
On its own, the benefits of this transformation are tremendous. Water can be recycled, nutrients don’t spill into groundwater, pests are minimized, land area is minimized. The consumers and farms themselves benefit as well, with production being much more predictable, so we can eat tomatoes even during winter. The next step in this type of transformation is to use AI to increase the density and wellbeing of the plants.
BlueRedGold was founded to deliver on a number of technological advancements aimed at increasing the efficiency of indoor farming. Firstly, we build robots that are specialized at handling very delicate plants and monitoring their health. For instance, we use hyperspectral cameras to understand the chemical composition of plants and the environment surrounding them. Secondly, and perhaps more importantly in the grand scheme of things, we use AI to optimize things. Now, it’s worth specifying exactly what we mean by both AI and optimization.
There is undeniably a revolution happening surrounding what is commonly referred to as “AI”. The term itself has existed for a long time and has applied to expert systems such as the one that beat Garry Kasparov in chess during the 90s. However, what AI must refer to nowadays is unquestionably state of the art machine learning, which is to say, deep neural networks (aka DNNs). One of the most fascinating abilities of modern DNNs is their ability to produce approximate solutions to problems that have been prohibitively expensive to solve historically. This is where the discipline of “optimization” enters the picture.
Historically, there have been practitioners of something called Operations Research (OR) whose job was to create algorithms that generate efficient decision-making plans. For example, an OR practitioner might write an algorithm that plans where airplanes fly to minimize costs while maximizing profits. The challenge has always been that these algorithms are not only difficult to produce but are also tremendously expensive to run. The more parameters, constraints, and combinations of decisions that are available, the more computers are required to produce good plans. What AI introduces, is the ability to generate very efficient approximate plans with extremely small computational costs. But, what does any of this have to do with indoor farming? Oh wow, let me tell you!
Large warehouses tend to be run by control systems that manage the state of items, personnel, and other facets surrounding day-to-day operations. Nevertheless, these warehouse management systems (WMS) tend to be fairly rudimentary when it comes to planning and opportunistically generate small picklists for people on the warehouse floor. Conversely, BlueRedGold’s indoor farm for saffron has a mind-boggling number of plants, trays, robots, and production lines that all can be monitored and steered individually. There are so many trays to move around, that it’s like planning the traffic for a small city. This is the first scenario where an AI-powered optimizer comes into play. By continuously generating optimal plans for what sequence of plants to move around, we can reduce operational overhead by over 40%. This is a game changer, because it effectively translates to an increase in effective growing area. We can essentially increase our production capacity and improve our service levels for free. That’s fairly impressive!
But there are more things that can be optimized which relate directly to the wellness of the plants themselves. For instance, we look at all parameter choices that affect time between harvests. This is not a vague analytical type of optimization (although we do that sort of regression analysis as well), but is about easy to quantify and discrete choices such as, how many flowers should we be picking per hour to increase the yield while minimizing harvest time?
In our view, AI is poised to transform the way we approach indoor farming. It has the potential to increase efficiency, reduce our environmental footprint, and usher in a new age of predictable, sustainable, and affordable food production. To learn more about the systems we have developed, feel free to reach out to firstname.lastname@example.org.