I've been playing with AppleML and GoogleML and neither seem ready for self driving cars to differentiate grandma or the school kids to feed their moral dilemma AI avoidance algorithm. But they can grab text and other objects.