Autonomous delivery is one of the most visible robotics ambitions and one of the slowest to arrive. Vehicles now operate without a driver in specific cities, on specific routes, under specific conditions, and the technology genuinely works within those limits. The gap between that and a van that turns up at any address is where the interesting engineering problems actually live.
The driving is the part that is closest to solved, and even there the difficulty is not the ordinary road. It is the unusual few seconds: a road closed with cones, a delivery driver double-parked, a hand signal from someone directing traffic, snow covering the markings. A human handles these by understanding the situation rather than by recognising it, and that is a different capability from following a road safely, which is why progress looks fast up to a point and then slows dramatically.
The larger obstacle is the last thirty metres, which is rarely discussed and rarely automated. A parcel delivery involves parking somewhere not designed for parking, finding a door that may be round the back, dealing with a gate, a buzzer, a dog or a neighbour, deciding whether leaving the item is sensible, and occasionally carrying it upstairs. None of this is driving, all of it requires judgement about a place the vehicle has never seen, and it is the majority of the time spent on each delivery.
Where the model does work is where those steps are removed. Fixed routes between depots. Movement inside a site, such as a hospital or a factory, where the environment is controlled and the destinations are known. Collection points where the customer comes to the vehicle rather than the vehicle reaching the door. In each case the automation succeeded by changing the task into one without the hard part, which is the same pattern as almost everywhere else in this field.
For a small business the near-term implication is modest, and worth stating plainly because the coverage suggests otherwise. Deliveries will keep arriving with a person for a long time. What changes sooner is the middle of the chain - the trunk routes and the depot work - which affects cost and timing rather than the experience at the door. Anyone planning around driverless delivery to their own customers is planning around the part that is furthest away.