This is a fun idea, but its harder than it sounds.
Pay attention next time you're in a grocery store. Damn near everything is different shapes and sizes. Sure, canned goods have *some* standardization, but there are no less than 3 styles of cans, (nesting, non nesting, and pull top) in a dozen or more sizes. And that's just canned goods, don't even think about things like sugar and flour, which come in fucking paper bags that can be punctured by a fingertip, let alone a robot arm. There are dozens of styles of salad dressing bottles, etc. My point is, having a robot stock/select these items from a shelf, or even pick single items from a case on a pallet is, at this stage, still not really that feasible. Quite a few advancements need to be made in robotic arms to be able to universally manipulate the erratic spectrum of shapes, sizes and weights of grocery items.
So, automatic delivery is one thing, what with your self driving delivery trucks and truck to door drones that everyone keeps raving about, but at the other end is still going to be some poor bastard rushing around a warehouse, throwing your selections into a box. maybe his job will turn into the guy who re-fills the giant soda-machine like dispenser that deals with *most* grocery items, and troubleshoots the inevitable jams it suffers from, but the level of automation this idea requires is an insanely complex system, dealing with literally hundreds of thousands of unique shapes, and this does not even address the varying degrees of durability. A plastic jar of peanut butter neatly handles a fall from 4 feet up onto a hard surface. A glass bottle of beer from the same height becomes a hilariously large mess.
Its a fun idea, and in small scale/limited selection (like any vending machine) its probably fairly operable, but to scale it up to grocery store level is an engineering nightmare/marvel waiting to be built.