In popular culture, autonomous vehicles (AVs) tend to be depicted as robotaxis traversing city streets, or self-driving cars racing down highways.
In fact, there are many other types of spaces where AV technology will be required — for example, privately-owned parking garages.
DeepMap has partnered with automakers around the world to build high-definition maps of complex, multi-level parking garages for automated valet parking (AVP). We’ve delivered centimeter-level-accuracy maps of the interiors, and helped customers align the maps to neighboring public streets.
We’ve also built maps of outdoor parking lots, which our CTO Mark Wheeler describes as “massive, complicated, unlabeled intersections.”
In today’s Here Be Dragons, we discuss autonomy and privately-owned spaces — parking garages, parking lots, private roads, driveways — all the areas that lie in between or at the edge of city streets, and state and federal highways.
Note: In each issue of Here Be Dragons, we highlight an ancient map from the David Rumsey Map Collection, located down the road from DeepMap at Stanford University.
Mapping the Private World
By Brad Templeton, Guest Author
- Most mapping for autonomous vehicles (AVs) has been of public roads, which follow a system of rules. But, driving requires navigating parking lots, garages, and private roads, which follow fewer rigid rules, and may be harder to access.
- Private parties, such as parking garage owners, will want self-driving cars to be able to enter their space. Tools are needed to make it easy for them to participate in mapping.
- Private spaces will leverage maps to perform active management of busy pick-up/drop-off locations at airports, arenas, and busy offices.
- Projects are already live that map parking garages so cars can valet park themselves.
Most of the effort in mapping for driving has focused on the public streets. This is where cars drive, and drive quickly, so they need highly-reliable data. The private world is also quite large, and lots of driving takes place there — on private streets, in parking lots, in driveways, and in pick-up/drop-off spots and lanes for large buildings and facilities.
Much of that needs mapping, but most of it has not been mapped. There are several key reasons for that:
- The benefits of mapping private spaces accrue mainly to those who use the space, in particular the property owner, so there is an expectation that they should bear some of the cost.
- Permission is needed to enter, and sometimes to photograph, private spaces. Many of the most important places are “open to the public,” but this still is an impediment to people wanting to do a commercial activity there.
- Private spaces do not need to follow many rules, and thus are harder to figure out with automated tools. There are no standards for lanes or signs, for example.
A self-driving car is a lot less useful if it can’t start and stop trips in private spaces. Being picked up and dropped off on public curbs works for some trips, but not for plenty of them.
Problem 3 is a particular challenge. Public streets follow rules, and there are laws demanding they follow these rules. This makes it easier to understand and map them.
A private parking lot might have a sign designating a parking space as reserved, or a lane as one-way, but there’s no standard way to do that. You may need human intelligence to understand some of the “rules” of parking lots and other private spaces.
It can even be argued that those who hope to make a vehicle which drives public roads with limited maps still need more detailed maps in a private area. Such cars seek to have perfect understanding of public roads in real time from just their sensors; the regularity of how public roads follow rules is necessary for this to even have a chance.
Early attempts at self-driving in parking lots without maps, even under human supervision, have not fared well, and they are not ready for real production deployment as cars make wrong moves and pause, blocking traffic. Trying to figure your way around a parking lot with no map data is a bit like solving a maze from the inside, compared to the birds-eye view a map can offer.
Fortunately, private property owners want people to be able to come and go in their robocars. We would all love to be able to summon our car while we are at the cash register of a store, and walk right out to it in the loading area.
If buildings have big parking lots or pick-up/drop-off lanes and zones, the owners want the cars to use them efficiently. Private homeowners want cars to be able to come up the driveway or even park in the garage. Gated communities want authorized people to be able use the roads there.
Private property owners may be more willing — and it may be necessary — to provide human input on mapping their spaces, to add that knowledge that only humans can figure out. They might also add rules that are only for robots, not for human drivers.
What is needed — nobody has yet built this — will be tools to make it easy and cheap for the owners and operators of private spaces to quickly build maps of their spaces. For outdoor spaces, it may be as simple as getting an aerial photograph and drawing lines and rules on it. For indoor spaces, it may involve sticking a borrowed portable mapping sensor suite on a standard car and driving around, then drawing the lines.
Pick-up/drop-off lanes and zones, as found at airports, big buildings, and public venues, have even more complexity than can be recorded in a map. Imagine trying to empty a football stadium after a game. A map can describe the rules and locations, but you may need more, such as a computer system handling traffic. The football stadium owners need not run this computer; they would likely contract it out to somebody who does this for many locations.
The map would contain one extra piece of important information — how to connect with the “traffic manager.” Using some agreed upon protocol, vehicles coming to pick up passengers would get a reserved spot and time to do their pickup, and would not enter the area until it was time. That would assure no traffic jams.
If very heavy capacity is needed, priority could be given to vehicles picking up many people. A stadium might arrange for buses to come to get 40 people all going north, and take them to a satellite pick-up spot somewhere to the north where they meet more private transportation. A stadium with room for 40 bus stops might drain people as fast as they can walk onto a bus, if the buses have multiple boarding doors. (Some would also take the buses nearly all the way home or walk or bike as they might today.)
We’ve already seen some experiments in self-parking in public lots. For example, at the Stuttgart Airport, you can pull up to one of the parking lots in certain Mercedes cars, step out, and have the car park itself.
The car has a map of the parking lot, and communicates with the lot to arrange parking. It drives through the lot to a special area, where it finds and takes a spot. Later, when the passenger lands, they use their phone to tell the car to drive back down to the staging area. In the future, cars will be able to get maps and communication information for many lots, and even park themselves “valet dense,” so we can park more cars in less space.
Whatever the system, there’s an opportunity for great improvements in transportation, including the part on private property, thanks to self-driving technology.
* * *
* * *
About DeepMap: DeepMap is accelerating safe autonomy by providing the world’s best autonomous mapping and localization solutions. DeepMap delivers the technology necessary for self-driving vehicles to navigate in a complex and unpredictable environment. The company addresses three important elements: precise high-definition (HD) mapping, ultra-accurate real-time localization, and the server-side infrastructure to support massive global scaling. DeepMap was founded in 2016 and is headquartered in Palo Alto, Calif., with offices in Beijing and Guangzhou, China. Investors include Andreessen Horowitz, Accel, GSR Ventures, Generation, Goldman Sachs, NVIDIA, and Robert Bosch Venture Capital. For more information, see www.deepmap.ai.