Not only will Waymo attempt to warm the public to self-driving cars with an ad campaign, Google’s sister company has put out a document explaining how the whole process works.
Waymo’s safety report is meant to give readers insight into how Waymo treats the idea of safety in an autonomous car. This includes how the cars are tested, how the systems operate in public and what steps Waymo is taking to specifically address the idea of passenger and pedestrian safety.
You can read the report in full in PDF form here, but if you’re looking for a TL;DR version, we’ve got you covered.
The idea of ‘safety by design’
Waymo uses the term “safety by design,” its mantra for how it approaches every step of the autonomy process.
Examples include behavioral safety, which involves the decisions the vehicles make while on the road, and functional safety, which ensures the vehicle works in a safe manner when system faults or other errors occur. In the case of the latter, there are redundancies like a second whole computer that can take over if the first one fails.
Of course, crash safety also factors in, and since Waymo is currently using a fleet of Chrysler Pacifica minivans, its fleet is already proven safe through standard federal crash tests.
Vehicles are taught some of these things through real-world testing, simulations and functional analysis, the latter of which involves a whole bunch of math.
How the car actually works
Waymo also gives some insight into the physical systems that help grant the vehicle Level 4 autonomy, which means autonomy that operates within given constraints but is left to its own devices.
Its fleet of Pacificas is outfitted with a lidar system that covers the front, rear and sides of the vehicle to map the world around it. Additional radar systems are at each corner to keep track of things within the van’s vicinity, and a set of 360-degree cameras atop the van give the computers a human’s-eye view of the world around it.
Waymo also touches on the software side of things. Its system is tasked with perceiving the world around it, predicting the behavior of everything in its area and subsequently planning a path down the road.
For the time being, Waymo’s self-driving minivans are constrained to what it calls an Operational Design Domain, or an area that’s been proven to provide for safe use of its autonomous systems. It will eventually expand that domain, but for now, if a user tries to input an address outside of that area, the van won’t go there.
Testing, testing and more testing
Waymo places its testing processes into three buckets — the car itself, the hardware and the software. The first portion is already taken care of, as the National Highway Traffic Safety Administration certified the Chrysler Pacifica for sale, and if it’s safe enough for public consumption, it’s good for Waymo.
The hardware is tested, well, just about everywhere. Waymo and FCA both conducted hardware tests on test tracks, in simulations and inside the lab. The tests cover not only the in-house hardware from Waymo, but things like door locks, doors, brakes and steering.
The software follows a three-stage testing process. Simulation tests allow the software to be fed information as if it were on the road, which is probably the safest way to start throwing random things at it. Some drivers will also test new software on Waymo’s 90-plus-acre private test track. Eventually, it’ll hit public roads for finessing and fine-tuning.
Using simulators might not seem that wise, as the real world is wholly random and generally without reason, but scale is the key. Every day, Waymo operates some 25,000 virtual autonomous cars over millions of virtual miles. Good luck trying to pay for that many cars and lidar systems with a corporate card, no matter if it belongs to Google or not.
Current semiautonomous systems have a hard time in inclement weather. Waymo knows this, and is attempting to ensure its system will have a leg up on Mother Nature. It bombards its hardware with ultraviolet radiation, big jets of water, ice, corrosion-friendly saltwater and hardcore vibrations.
Dealing with Jane and John Q. Public
Finally, Waymo outlines how it works with the general public to refine the side of the car the consumer will see. Right now, Waymo’s vehicles offer a display that lets riders see what the car sees, whether it’s mapped elements such as roads or discovered elements such as cyclists.
Of course, there’s an app involved. People involved in Waymo’s first public trial in Arizona can order a car with Waymo’s app, which can also be used to start the ride once everyone is inside (there’s also a button for the car in this). If someone needs an unintended pit stop, there’s a “Pull Over” button that will signal the vehicle to find the first safe location to pull over.
Waymo also has a support team that can be reached by phone, app or a button inside the car. Whether riders have general questions or an emergency, phone-based support is there to help or educate as necessary. Knowledge is power, after all.
When it comes to users with disabilities, Waymo is working on several solutions. Its mobile app can be accessed by voice commands, for example. The company is looking at adding features to its vehicles so visually impaired users can locate the vehicle once it’s arrived, in addition to adding audio cues to its app so they know where they are at all times. The buttons in Waymo’s minivans have Braille, too.
The cars are also engineered to understand and react to emergency vehicles as necessary. Once an ambulance, police car or fire truck has been detected by light patterns or audio, the vehicle will find a safe spot to pull over. It can also merely yield, in the event of a divided highway or other situation that doesn’t require a full stop.
And that’s about it! Waymo’s report is more than 30 pages, so yes, that was a TL;DR.
If you’re still hungry for more, check out Waymo’s full safety report, or just move to Arizona and attempt to worm your way into the pilot program. No guarantees the latter will work, though.