How Self-Driving Cars Can Figure Out Parking - Forbes
Robocars are very close to being able to drive the roads at a safety level to match humans. But what about dropping you off, picking you up, and parking or waiting?
For most teams, driving on the roads will be done with detailed maps of the road. They will know where to go on any road, and also how to tell if the road differs from the map and what to do in that situation. (All cars will have at least rudimentary ability to drive without maps for this situation.)
Those maps will include the locations of parking spaces and standing spaces on the public roads. Standing spaces can include the entrances to driveways and spaces in front of fire hydrants, suitable for pick-up an drop-off. Cities have long allocated "taxi stands" for waiting, as well as pick-up/drop-off areas, and recently I have observed such areas being allocated for use with Uber/Lyft style services.
Drop-off points and parking spots are very commonly on private land, though. How will cars know where to go if you are heading to a private destination?
For important private venues -- schools, public buildings, and large buildings, the mapping companies will map them just as they do roads. Most of them would allow a human-driven mapping car to pay a visit. In addition, privately driven cars with sensors will also frequently travel to them and can gather mapping data.
Generally, such maps need a basic level of human review, and as such, buildings below a certain threshold of traffic will not get mapped as quickly as high traffic buildings. This can be solved by allowing the owners, and even the public, to map these spaces. Somebody motivated will generally be able to start with an aerial image of the private area to mark out the geometry of the space -- where can cars drive, what is sidewalk, where any pick-up/drop-off points are and, as described below, waiting/parking areas. If there are not high-quality aerial image of a location (ie. it's non-urban) owners could, at a pretty low price, request a drone overflight to help.
The cars don't use the aerial imagery to navigate -- this is just to get the rough shape. A visit by a robocar (or human-driven car) with the right sensors is needed. That gets easier and easier as more robocars are available. Such cars can have a remote human supervisor when they enter the unknown space to map it. The supervisor almost never "drives" it like a human with a wheel, rather they give strategic commands, like pointing to a spot on an image and saying to drive to that spot. The car worries about not hit anything along the way, or noticing things the remote operator might not see.