The detailed report looks at how the systems in Waymo cars operate, the type of data they record, what happens when something goes wrong and how the car deals with an accident.
The submission also pulls back the curtain on hardware testing processes, and how passengers might interact with the car from the cabin - but perhaps most interesting is the detail about how self-driving cars interpret, analyse and act on information about the world around them.
According to the document, self-driving cars need to first have a detailed understanding of where they are in the world. Rather than relying on GPS data, Waymo cars combine data from their sensors with in-built mapping information for a location.
The sensor array – which includes radar, camera and Lidar units along with supplemental microphones to identify sirens – then scans the road around it for other vehicles, people or obstacles, before trying to predict what each of them will do.
Based on all that information, the car plots a course through the chaos. It’s worth remembering, all these calculations need to happen flawlessly, thousands of times every second.
If one of the steps in the process doesn’t work, there are a number of redundancies built into Waymo vehicles. There are backups for all the major systems, including steering, braking and vehicle positioning sensors.
Should something go wrong, the car is always logging data, allowing lessons from one mistake to be applied across the entire vehicle fleet.
Cars are equipped to know when they’re involved in an accident, and automatically notify Waymo HQ so the team can contact the relevant authorities. If there’s someone in the car, they can communicate directly with the Waymo team after the crash as well.
As you’d expect, every crash is analysed and appropriate tweaks are made to make the same problem (hopefully) doesn’t occur again.
Along with its real-world testing, Waymo is constantly running its software through virtual tests in a homemade simulator. The company says up to 25,000 virtual vehicles cover up to 8 million miles (12.8 million kilometres) in simulation, fine-tuning existing skills and practising unique or quirky situations rarely encountered in the real world.
Above: a... slightly modified Waymo image by CarAdvice, highlighting just one of the issues autonomous cars will need to deal with
All of this work is, according to Google, necessary to cut the global road toll and mobilise people who are otherwise unable to drive themselves.
The report says 94 per cent of all US accidents are caused by human error, while more than 1.2 million people were killed in road accidents in 2013. Annually, the company says road trauma costs US$594 billion ($759 billion).
Self-driving cars have the potential to slash the global road toll, saving countless billions of dollars in the process. By taking over driving duties, autonomous vehicles also promise to turn the 42 hours most people waste sitting in traffic every year into more productive, enjoyable time.
Waymo is the first company to voluntarily submit a safety report to the US Government, after it suggested last year (and again last month) that firms developing autonomous vehicles should share their safety protocols.