The Future is Now: Advances in Agricultural Robots

These days, it’s rare to find an industry untouched by artificial intelligence. From the machine learning that predicts customer behaviors and purchasing patterns to the machine vision that drives facial recognition technology, everyday life is filled with examples of algorithms and automation at work.

Even the hardware has become somewhat ubiquitous. Robots assemble cars for the automotive industry, assist surgeons during medical operations, vacuum floors for busy homeowners. While some industries like agriculture are generally slower to adopt new technologies, big changes are afoot.

Agricultural robots are already in operation on farms around the world, disrupting the industry in new and different ways. Dr. Khasha Ghaffarzadeh, IDTechEx Research Director and author of “Agricultural Robots and AI: A Question of When and Not If,” discusses some of the futuristic things he’s seeing right now.

Q: What are some of the most interesting ways you see robots being used in the agriculture industry?

Dr. Ghaffarzadeh: There are multiple strands of development. In one strand, companies are trying to develop a highly intelligent tractor-pulled implement. These implements are vision-enabled, meaning that you have a deep-learning-based vision algorithm that can see and identify and detect different types of crops. Therefore, the implement is able to take a very precise action. Here, the emphasis is primarily on the vision system and the ultra-precise action, whether that action is a precision spraying system, a mechanical process or something else.

Another manifestation of this is to have a precision system with advanced vision technology together with some sort of autonomous mobility technology. These are generally small- to mid-sized robots that are equipped with a vision system and can move around more or less autonomously.

The third major strand is to try to develop robots that can pick different kinds of fresh fruit. There are a lot of advancements in terms of the design of the end effector, the design of the machine and, crucially, the development of all the algorithms that are used to detect the fruit and localize the robotic implement to pick it.

Q: What has kept agricultural robots from disrupting the industry sooner?

Dr. Ghaffarzadeh: I think a number of things have taken place. One is that the technology has finally been demonstrated. There have been some advances in technology in the past seven or eight years that have really enabled this field to come about. Without these advances, particularly in autonomous mobility and vision technology, none of this would be possible. There have been some technological prerequisites that have now been fulfilled. From now on, we don’t necessarily need technological breakthroughs or radical innovation to further enable this process. We just need incremental gains.

And of course, these things take time because you need to develop a machine, the first and second generations of which are generally prototypes. You need to develop an algorithm and have a lot of field experience on the farm to get a lot of feedback about your robot to improve the machine and the algorithm. Then, you need to go through this journey of trying to ruggedize your robot and make it reliable. Now, you need to think about the business side of finding customers, demonstrating the use case and value proposition, finding the early adopters, etc. All of this takes a lot of time, but a lot of the fundamental hurdles have been cleared away. Now, it is simply a question of when.

Q: Why is machine learning so essential to a robot’s success? How is the technology being improved for the future?

Dr. Ghaffarzadeh: For a robot to be able to take an action, it needs to be able to see, just like a human being. In the past, it was very difficult to use classical computer vision technology to teach a computer to identify different types of crops. Now, it isn’t easy, but at least there is a pathway to be able to do it. In all of these applications, the emphasis is on seeing something in a very precise way, localizing it, detecting it and taking a very precise action. This applies to weeding applications, robotic seeding applications and picking fresh fruit.

Many of the advances that have improved machine vision technology have not been driven by agriculture, but agriculture has benefited from them. Object recognition, which allows multiple classes of objects for example cats and dogs to be identified and localized with a bounding box, is one example. The algorithm for this is very good and can be used in agricultural applications to identify different crops. This is different from something like autonomous driving because, in this case, you need to be able to perceive the world as well. Mistakes in this arena can be fatal, so you need to spend a lot of time on it, gathering data, training and testing the algorithm, and so on, especially to ensure that you overcome the rare edge cases.

I think that one of the key trends going forward in machine vision technology is that these very heavy algorithms will become computationally light. This means it will be easier to train the algorithm and to run it on the robot itself, which should drive down the cost of the entire system over time.

Q: How do you see farmers successfully integrating these technologies into their workflow?

Dr. Ghaffarzadeh: In the short term, humans will remain in the equation. All of these robots require expert operators, and they need someone around to quickly troubleshoot issues that arise. The human may not need to be right next to the robot, but there will be some level of supervision going on.

When you think about the adoption of these technologies by the farmers, it’s important to look at the business models of some of the companies. Some of them are saying, “Maybe we shouldn’t sell this equipment like we would sell a tractor.” Perhaps, the equipment requires expert input, or the farmer doesn’t know how to use it yet, or the machinery is very expensive. What these companies are saying is that instead of sending the farmers these machines, they will offer it as a service. The company will come in and do the weeding, for example, using the agricultural robots and charge the farmer for this service based on how much of the field is weeded.

This allows the companies to gather more data and images, helping them to improve their robots all the time. They get instant feedback about the design of the robots. Whenever a new technology enters the marketplace, there’s also this fear that it will soon become obsolete. We are in a time when technology advances very quickly, so if you offer this as a service, you de-risk that a little bit for the end user.

Q: When do you expect to see agricultural robots on farms become the norm?

Dr. Ghaffarzadeh: Let’s go through the different categories. Tractors are already able to operate at level four autonomy (driving around without a human but only in a specific area), and they are very popular in places like California. I think this autonomy will go further. That said, going from level four to level five autonomy (driving around without a human anywhere) is not yet completely economically justified.