How to retrive camera pixel data?

I have multiple objects on the screen, each one looking in different directions. They will have an artificial neural network attached to them, which make decisions about their actions based on what the objects "see". To do this I need to be able to get the raw pixel of the scene from the agents perspective.

I would think the best way to do this would be to have cameras attached to the agents. But is there a way to retrieve the rendered image of the cameras view in the script?