Automating agriculture

In the so-called “Age of Automation” that we live in, breakthroughs in big data and machine learning have set off waves in a variety of industries. One of the more unexpected and interesting fields that has begun to exploit automation is agriculture. Professor Ken Goldberg, an expert in robotics and AI at UC Berkeley, is using imaging data collected from flying drones over fields to train a machine learning model that determines water needs per plant, which then instructs robots to actually dispense the water.

Goldberg first started exploring the potential of automation technology in agriculture a few years ago when he collaborated with professors at UC Merced and UC Davis as a part of CITRIS, the Center for Information Technology Research in the Interest of Society. At the time, the California drought was extremely severe, prompting Goldberg and his colleagues to urgently search for a solution.

Upon identifying a major contributor to the drought, the amount of water wasted in commonly-used irrigation systems, the CITRIS team developed a solution to reduce water usage — the Robot-Assisted Precision Irrigation and Diagnostics System, or RAPID system in short — which builds upon existing drip irrigation systems.

To provide some context on modern irrigation, there exist three main types of irrigation techniques: surface, overhead, and drip.

In surface irrigation, gravity carries water through furrows and basins dug across the field. Also known as flood irrigation, this method is the most commonly used globally. However, it easily falls victim to sources of water waste like evaporation from exposure and runoff from oversaturating soil.

Meanwhile, in overhead irrigation, pressurized rotating sprinkler devices found on the vast majority of grass lawns spray mist over plants within their radii. This wastes less water than surface irrigation, but still suffers from evaporation and runoff.

The last main technique is drip irrigation, which delivers water drop by drop to the roots of plants, often just below the surface of the field. Thus, in theory, drip irrigation maximizes plant hydration and minimizes water loss. In particular, the slow, consistent rate of water flow and the lower elevation of water placement drastically reduces evaporation and soil oversaturation.

In pursuit of Goldberg’s self-stated goal “to tune the amount of water to the individual plant,” he and his team first were inclined to use remote electronic valves farmers could control for each plant, but soon realized that they were too expensive, especially when “the farmer usually makes only a few pennies per plant.” Such remote valves are also prone to error, further driving up the cost.

While drip irrigation initially appears to be the best solution to conserving water, drip irrigation systems in practice clearly runs into some issues. This led Goldberg to turn to the RAPID system, the next stage in the advancement of classic drip irrigation techniques. RAPID uses cheap passive valves which can be manually twisted to regulate water provisions on an individual plant basis. The valves are paired with emitters that indicate how much water is needed for each plant.

In order to make the RAPID system viable, Goldberg had to overcome two major challenges — first, the amount of labor required to twist the valves on thousands of plants in a single field, and second, determining the water needs of each individual plant accurately.

To address the first challenge, Goldberg and his team constructed low-cost robots that could take up the exhausting work of checking emitters and manually twisting valves for each plant. However, in order for the robots to know the right amount to twist the valves, the model Goldberg was working on would need to determine how much water each plant needs, the second challenge.

An example of a drone that could gather image data.

To do this, CITRIS would first fly UAV drones over fields to gather extensive image data. After that, their machine learning model would process that data to statistically determine significant features, such as the drainage of the soil, the angle at which the plant is growing, the frequency and strength of wind the plant generally experiences, and the amount of shade affecting the plant growth. Then, the model would combine these metrics in an optimization function to evaluate the ideal amount of water for each crop.

This combination of robot valve twisting and model prediction directly helps reduce the problem Goldberg and his team initially set out to solve — reducing water waste. However, this also opens the door to another benefit, the stimulation of crop yield.

Now, individual plants are receiving water based on their particular needs rather than receiving a random amount based on the person giving water on any specific day. With better-targeted water allocation, crops have a better opportunity to grow. In an even more exciting potential application, Goldberg spoke to how this robot-model system doesn’t have to stop with water allocation. With a few changes and more imaging, the same system could be used to even detect plant disease and dispense pesticides.

Despite these benefits of introducing AI to agriculture, a common fear voiced globally is that the quick move to automation is going to replace jobs and drive up unemployment. Goldberg makes sure to quell concerns about disruption to the job market.

He notes, “Historically, machines have taken over agricultural jobs, but it’s also important to note that there are many new jobs created. If you look through history, technology does disrupt some jobs, but it also does create new jobs.”

Long story short, Goldberg is optimistic for what the future holds for the introduction of AI to agriculture, and is not dissuaded by the somewhat bumpy transition that workers in the field may experience.

As two students at UC Berkeley studying computer science and engineering, we could not be more fascinated and excited by the way that rather abstract concepts in math and statistics are making a practical difference in one of the most fundamental aspects of life — food production.

The Food and Agricultural Organization of the United Nations notes that more than 820 million people do not have enough food to eat around the world. In addition, it found that while the number of people without secure access to food has been declining year to year for decades, that trend is no longer the case today. Automating agriculture has the unique potential to help remedy this problem by better allocating limited resources for food production to feed the world’s hungry. The work being done by Professor Goldberg is revolutionary in the scope of human lives it can affect. Though it will be an iterative process that is not entirely smooth sailing, the move to AI in agriculture will positively change the world.

The work that engineers do shapes the world around us. But given the technical nature of that work, non-engineers may not always realize the impact and reach of engineering research. In E185: The Art of STEM Communication, students learn about and practice written and verbal communication skills that can bring the world of engineering to a broader audience. They spend the semester researching projects within the College of Engineering, interviewing professors and graduate students, and ultimately writing about and presenting that work for a general audience. This piece is one of the outcomes of the Fall 2019 E185 course.

Master of Engineering at UC Berkeley with a focus on leadership. Learn more about the program through our publication.