Google recently began testing their self-driving car in the metro Phoenix area… Yesterday, representatives from Google met with some members of the Phoenix area bicycling community.
CAzB Leadership and Friends,
This was a fascinating meeting…Where Google is headed with their technology has tremendous potential to save lives and reduce injuries, including bicyclists. Let me try to convey the key points we registered, but first I’ll tell you who attended:
· Bob Beane, CAzB/PMBC
· Linda Torres (my better half, a cyclist and a good listener)
· Joe Struttman, Pecos Action Group
Google and related
· Jennifer Haroon, Leader of Business Operations for the Google Car Project
· Angie Welling, External Affairs/Public Policy for Google
· Gerardo Interiano, Google (will be taking over the region including AZ for the project)
· Rob Nash, Consultant to Google
· Brian Murray, Consultant to Google
The Goal of the Project
· The overall goal of the project is to come up with self-driving car technology that is much safer in operation than human-driven vehicles.
· An ancillary goal is to provide accessibility and mobility to the subset of people who cannot physically or safely drive.
· In development, to this point, an overriding principle is “conservative driving”…more on that below.
· As expected, despite our repeated probing, Google representatives would not reveal the “Secret Sauce” recipe for how all the technology comes together and how algorithms (or whatever) ultimately make driving decisions, however…
· Various technologies (camera, radar, lasers, sensors, GPS mapping, etc.) are used in combination to provide data input on everything going on within about 200 yards in all 360 degrees around the vehicle as it operates. The vehicle, with all technology on board (not reliant on an Internet connection), is self-sufficient in capability and can “see” things that human drivers may not be able to see. I did not hear anything specific related to night driving (e.g. no Infrared), but it seems that the car may be even relatively more safe at night.
· Data is used in multiple ways, including maintaining positioning, monitoring stop lights, identifying moving objects and hazards, collision avoidance, etc.
· The vehicle also is using data to make predictive projections about speed, direction, etc., of other objects within its field of view, not restricted to the roadway (i.e. any object within the 200 yard circle is literally “on the radar”, whether on the road, on the sidewalk, emerging from a parking lot or driveway, etc.
· “Machine learning” is being used to “train” vehicles to recognize scenarios, though the Google people repeatedly stated that the response is not a “hard coded” response, but rather somewhat (my term: artificial intelligence) an “educational” addition to possible outcomes…Again, part of the “Secret Sauce”.
· Known traffic laws (e.g. three foot safe passing distance) are being incorporated into the knowledge base of vehicles, and Google has even patented technology for recognizing bicyclist hand signals.
· There is redundancy built into the car in case of a tech failure. There is also a protocol for “confusion” or perceived unsafe conditions. If the vehicle becomes immersed in a situation that seems “confusing” it will first slow down and attempt to pull over. If that isn’t possible or there isn’t enough time, it will simply attempt to stop. It does sense both hazards in front and approaching from the rear (in other words, in a rapid stop situation it will try to avoid collisions both with the vehicle in front AND the one behind…how they do this was part of the “Secret Sauce” or something their attorneys don’t want them to discuss).
· As mentioned above, the cars are being designed to drive “conservatively”, including a 1.5 second delayed start at intersections.
· Having said the above, the cars are also designed to make right turn on red AND they have programmed a “human” factor of inching forward until space is clear specifically to let vehicles behind know that the car is still functioning and “knows” what to do (purely programmed in reaction to behavior of human drivers).
· There is said to be some prioritization of vulnerable roadway users in collision avoidance, though specifics weren’t provided.
· The Chandler, AZ, tests will be conducted with trained test drivers on board, however their intervention is expected to be absolutely minimal at this point in the development of the car.
· Crash data to date: More than 1.5 million miles driven, 19 crashes, most caused by other drivers rear-ending the Google cars.
· Testing: 3,000,000 miles of simulated driving can be done, DAILY. Data from on-road driving is uploaded and even when drivers take over the car Google can simulate what would have happened if they did not.
· We saw a very detailed PowerPoint presentation with moving simulations. While those don’t appear to be available on line, there are videos and some good information about the program at www.google.com/selfdrivingcar/ .
Please feel free to send any questions. Joe Struttman may have some additional comments, as well.
This briefing and Q & A was well worth the 1.25 hours spent, and the Google folks were definitely taking notes on our questions and concerns. We did mention the Valley’s arterial grid structure with high-speed, multi-lane arterials with shopping/retail (with additional driveways) located near the major arterial intersections…and the fact that serious crashes frequently happen there. They mentioned that they also have “taught” their vehicle to try to avoid being in blind spots. So, we’ll see how it goes.