3 Ways Cruise HD Maps Give Our Self-Driving Cars An Edge
Especially when it comes to complex urban environments like San Francisco.
Written by Erin Antcliffe, Senior Product Manager, Mapping
At Cruise, we’re on a mission to build the world’s most advanced self-driving vehicles to safely connect people with the places, things and experiences they care about. The potential to save millions of lives, reshape our cities, give people more time, and restore freedom of movement for many motivates our teams at Cruise everyday.
My team, the Mapping team, creates maps that store important pre-computed information about the world. This allows our self-driving cars to focus on other roadway users.
High Definition (HD) maps serve as one part of the self-driving car’s “eyes”. HD Maps are made up of two main types of assets:
- 3D tiles that are rendered from LiDAR sensor data
- Semantic labels that are applied on top of this sensor data to give meaning to the environment
HD maps provide self-driving cars with information about real-world environmental features, such as the boundaries of lanes, location of traffic lights, and presence of curbs at the edge of the roadway. All of this information lightens the real-time processing load on the self-driving car, letting it focus on navigating around dynamic actors in the roadway, such as vehicles, bicycles and pedestrians.
As the autonomous vehicle industry has grown over the past few years, companies have had to decide whether to create and maintain their own HD maps or use maps from one of the many HD mapping companies that have evolved to support the industry.
At Cruise, we’ve decided to build our own maps to maintain full control over the format, quality, and taxonomy. This allows us to operate more successfully in complex urban environments, keep our maps up-to-date with real-world changes, and facilitate rapid self-driving development.
Precise maps help self-driving cars localize in complex urban environments
In today’s smartphone era, you’ve likely used GPS on your phone to find out where you are and where you’re headed. If you’ve tried this in a city as complex as San Francisco, you may have encountered an inaccurate location reading that tells you you’re several blocks away from where you actually are. This is because tall buildings like the Transamerica pyramid obscure the line of sight between your phone and GPS satellites, leaving you with low confidence in your position (this is known as the urban canyon effect).
This inaccuracy might be fine if you are able to figure out your location relative to where you’re trying to go on foot. But for a self-driving car, this low level of positional accuracy is paralyzing.
To solve for this, self-driving cars use LiDAR sensors to compare their surrounding environment with the 3D map and determine their location down to centimeter-level accuracy. This allows us to perform sensitive maneuvers with a high degree of confidence by enabling the car to make sense of its surroundings and plan it’s next action accordingly, such as moving past a bicycle in the next lane or anticipating pedestrians crossing the street at an upcoming mid-block crosswalk.
Updating maps with San Francisco’s constantly changing streets
As I described above, HD maps provide a critical input to self-driving cars. Given this fact, it’s important to keep the map up-to-date with changes out on the road. Dense urban areas like San Francisco are constantly undergoing construction projects like adding protected bike lanes, equipping better traffic control signals, and building new areas of residential or commercial development.
We have developed sophisticated product and operational solutions to detect real-world changes and send map updates to every autonomous vehicle in the fleet in minutes.
This is one of the reasons why maintaining our own maps is a competitive advantage.
Facilitating rapid self-driving development
Another major advantage to producing our own maps is the ability to leverage our map production platform to experiment more quickly on cutting-edge autonomous feature development.
Just like human drivers, self-driving cars perform better when they are familiar with the road environment and expected behavior of other vehicles (e.g. encountering an all-way stop intersection). We can give our fleet an edge by using maps to encode this information based on thousands of interactions in a given area. As we experiment with different ways to represent spatial data, we can quickly test our hypotheses, iterating on different versions of a new map feature to assess its impact on our self-driving car’s performance. Once we find the right solution, our map production system allows us to quickly scale the new feature across the entire map.
By working closely together with Cruise Engineering teams to develop map features side-by-side with self-driving behavior improvements, we can more quickly deliver on our vision of a fully self-driving car.
Cruise maps are a foundational advantage to self-driving development
With the wide variety of challenging scenarios in San Francisco’s complex urban environment, HD maps are a tool that give self-driving cars an edge when driving. At Cruise, one of our fundamental advantages comes from producing our own HD maps using precision LiDAR and semantic mapping techniques. By building maps in-house, we maintain full control over the maintenance strategy and can quickly iterate on new map features to help self-driving cars get on the road more quickly.
Overall, Cruise maps empower our self-driving cars to drive in SF so we can achieve our mission of building the world’s most advanced self-driving vehicles to safely connect people with the places, things and experiences they care about. The sooner we deliver all-electric, self-driving cars, the sooner we can save lives, reduce emissions, take back our time, and democratize transportation.
Join the self-driving team
See you on the road!