I took a couple of hours and wrote a FruitBot.

The fun part was the model, basically it was inspired by the SVM method of solving classification problems -- where you create a hyperplane in R^N-1 to separate data points in R^N (ideally, all the 0s will lie on one side of the plane and all the 1s on the other).

In this case it took the form of finding the line that best divided my robot from the other robot (with the most attractive fruit set on my side) and chasing the fruit on my side of the line. The basic prototype seems to have achieved a reasonable amount of success, though I could definitely put a lot more work into it.

fruitbots.org/users/1678