dfw_gis wrote:What I enjoyed reading about was the separation of the stacked data layers incorporating and building on the base map being the 3D data, taking in local controls (striping, signals, signs, etc), and their approach to temporal issues throughout the day and week.

Right? Pre-computation mapping will still need human contribution, which should be exciting for everyone who likes to edit maps

You are correct, that is a very interesting article. Thank you for sharing. The link at the very beginning going to the wiki page comparing the various map services was a helpful bookmark too. What I enjoyed reading about was the separation of the stacked data layers incorporating and building on the base map being the 3D data, taking in local controls (striping, signals, signs, etc), and their approach to temporal issues throughout the day and week.

Absolutely! A strong foundation is a must to support everything being built on top of it Without the base being constantly revised and refreshed data-driven aspects that follow will not be at their maximum value and potential.

I sometimes wonder if the day will come to where we can see a seamless hand-off from the vehicle based navigation system to a pedestrian augmented reality view. A doorstep, driving, parking, walking, arriving scenario. Purely curiosity if the fusion will occur someday

You make a great point. I've seen other companies like Here technologies presenting articles of like-minded collection processes. I haven't heard of fleets adding collection features but not to say they haven't. On a different vector, I've had a conversation with a geospatial peer at Union Pacific and they outfit their locomotives with LiDAR collection devices.

This is an excellent illustration that I think is usually overlooked - most of the coverage of autonomous vehicles is of the vehicle hardware itself, but this shows the collected data and processing is a huge part. Companies such as Lyft and Uber already have a huge advantage in collecting map data - they've got the wheels on the ground. I could imagine them in short order placing devices on all their drivers' cars to collect the data the will eventually put the drivers out of business. Companies like Lyft could also find themselves owners of a huge repository of map data and routing services that could be licensed to 3rd parties - automakers make the hardware, Level5 provides the backend data & services. I wonder if UPS & FedEx are doing anything in this field too - think of their huge fleets and their delivery route processing. They could layer on cameras and LIDAR onto their fleets and provide services too.

And that does come back around to us - I wonder how Alphabet will make use of the work we've done - and will we be asked to do more?

A little late to the party, but folks here should also check out the company Nexar who uses its network of dash cams to collect data, and has just recently started using the data to produce information. Things like determining and predicting light timing, or recognizing and automatically warning about work zones.

Late to the discussion, but some thoughts:

There are two distinct strategies for autonomous vehicle's navigation logic: hardware-centric and map-centric.

Tesla is an example of the first: while the general drive routing comes from the normal GPS mapping, all the fine details of the drive depend on what the car senses in real-time. The map can be inaccurate without changing the safety of the drive. Speed limit change? New stop sign? An intersection replaced by a rotary? The car's internal capabilities can detect and manage those and other conditions.

Google's efforts in this area are more focused on having a super-reliable map and much less smarts on the vehicle itself. This approach requires the highly detailed, always updated map discussed in the article.

To my thinking, Tesla's approach will dominate. Distributed computing is more resilient, reliable and scaleable than centralized computing for this purpose. It's also not realistic to think that maps of even 95% of roads in developed economies can be maintained to the centimeter-level precision required by the centralized approach--let alone getting to 100%. If cars need to be smart enough to manage the 5-10% of poorly mapped roads, they're then smart enough to manage 100% of the roads.