Last year, in June, I published a news of a robot that can pick up fruits, as small as raspberry, automating the harvesting. Now I have stumbled on a news of a drone/robot that can fly in orchards to pick up fruits.
The news is impressive on several levels:
- flying in an orchard is tricky, and it is not just flying, the drone needs to get close to the fruits avoiding getting stuck in leaves and branches
- picking up fruits requires, first, to identify the fruit, evaluate its ripeness, work out a procedure to detach it from the plant with no damage to the fruit and minimal impact to the plant
- working out a pick up strategy minimising flying time
All of the above requires quite a bit of awareness on the drone side and this is where artificial intelligence steps in.
Notice that what seems trivial to us, like detecting a fruit and evaluating its ripeness is actually quite complicated. More than that. The evaluation of ripeness that I can give is way different from the one of a farmer. Looking at a yellowish apple would probably give me the idea that it is ready for harvest. That same apple would send a different message to a farmer who might, through experience, know that it is, yes, yellowish but that is only on the line of sight, it is actually in the shade and the other side would be greenish, indicating that a few more days are needed before picking it up. This kind of reasoning, based on experience, is what the drone needs to make, even more important for a drone since it willl not have the option of handling the apple, turning it to look at the other side.
However, this is not what happens. The intelligence is not in the drones but in the cart (shown in the photo) accompanying the drones. The drones have on board sensors that communicate with the cart where data are processed and commands issued to the drones. A single cart can manage several drones.
As shown in the photo, the robots are operating in orchards that have been grown to meet industrial farming criteria, hence the fruits are conveniently spaced and, basically, on a single vertical plane, making detection and harvesting easier.