Between 2017 and 2020, our virtual testing generated the equivalent* of over 3 billion real-world miles. By June of this year, we had exceeded 5 billion as our enhanced virtual development tooling and expanded team allowed our engineers to chew through an average of over 22 million miles each day. By year’s end, we expect to have driven the on-road equivalent of over 9 billion miles.
Another way to think about it: an on-road fleet of 100 trucks gets you ~100 round trips between Dallas and Houston each day, if you’re lucky. This fleet experiences only the driving scenarios and environmental conditions that happened to occur on that route, that day. In contrast, our virtual fleet drives the daily equivalent of over 46,000 round trips between Houston and Dallas, experiencing a diversity of scenarios and driving conditions that would take even a large on-road fleet decades to experience. Put another way, our virtual fleet “drives” the equivalent of 47 round trips. To the moon. Every day.
Two: Adding key capabilities and defining progress
The ability to drive over a million virtual miles a day has led to an explosive growth of technical progress, especially around key capabilities for highway and urban driving. To track our progress, we compile key metrics to understand performance trends at multiple levels of detail, and we weigh offline test results by exposure probability and severity. We use these metrics to make informed decisions and prioritize the next steps in the development of the Aurora Driver.
As we look to the next year, our progress puts us on track to improve our software dramatically every quarter, even as we expand our capabilities to include “long-tail” events, like navigating construction zones and debris in the road, detecting small objects, and accommodating emergency vehicles.