I see! But how do the agents see the diffused pixel data in order to decide whether or not to turn left or right? Reading back the image data from the GPU is very slow, at least from the method I tried using Canvas:newImageData()
