Agriculture as a testbed for robot learning.
Agriculture is a great testbed to apply robot learning on. Following are the three reasons agriculutre is a great testbed for Robot Learning.
Reason 1: Demand for labor in agriculture is high. Currently US has a major labor shortage. At this point its not even about the labor cost, but labor is simply not available. Farmers actually badly want the robots in their farms. This is unlike home robots, where people are not desperate for the robots just yet. Its more of a luxury than a necessity(at least for the majority of the people). To be fair, home robots are a necessity, especially for the elderly and disabled and that is an important problem too. But my next point will make it more clear why agriculture is a better testbed than healthcare.
Reason 2: Robots dont need to be around humans for most agriculture tasks. This is a pro for agriculture, since most of the robots currently are not safe to use around humans. If you want to deploy machine learning based robots in your home, you have to stress test it extremely well to ensure that it doesn’t hurt anyone. This is not needed for agriculture, since most of the time, the robot is not around humans. What’s the worst that can happen? Plants are bruised?
Reason 3: Food supply is critical for humans. The problems in this space will get more attention from the governments. This is unlike home robots, where the problems are not critical enough to get the attention of the government. The problems in home robotics are mostly a luxury problem, which is great that we are solving them, but they are not critical problems for the society (at the moment).
These are all great reasons in favor of agriculture being a good testbed for robots. However, there are some major challenges as well that I will mention briefly.
Challenge 1: Lack of data. This is a major challenge, since we don’t have a lot of data for robots in agriculture. Most of the SOTA vision models for example, don’t work off the shelf for field settings.
Challenge 2: External conditions are rough. Robots will be deployed in the field, and field conditions have to deal with sunlight, rain, snow, and other external conditions. These conditions are tough not only for robots, but for researchers working in this space as well.
Challenge 3: Hard to build simulation environments for agriculture. How do we simulate the gentle contact the robot makes with the leaves of the plants?