The technological foundation accelerating development of the Aurora Driver

At Aurora, we recognized early on that upfront, foundational investments in key technologies would enable us to build a safe and scalable Aurora Driver. Aurora Illuminated was a showcase of this approach—what we call self-driving 2.0.

In the videos below, Aurora’s engineers share how our data drives our decision-making and forecasting methods, the importance of our investment in sensor technology, and more. It’s these cutting-edge technologies that come together to expedite expansion of the Aurora Driver’s on-road capabilities and ready our vehicles for our planned deployment in 2023.

High-performance hardware: Purpose-built for reliability and commercialization

Our high-precision self-driving software can’t perform at its best when run entirely on off-the-shelf parts. Rather than relying on publicly available hardware, our tightly integrated hardware and software teams jointly design high-performance, custom-built components that meet the specific needs of the Aurora Driver and our business model. Check out this video for an inside look at our approach to hardware.

Our state-of-the-art perception system provides the Aurora Driver with advanced situational awareness, made possible by the integration of deep learning and probabilistic state estimation. Here, the Aurora team explains how our perception system fuses sensor data and other data inputs to capture all the significant details of the Aurora Driver’s environment.

The Aurora Atlas, our HD mapping system, enables smart, reliable autonomous driving. In this presentation, Aurora team members share how the safest drivers are the ones who not only know where they’re going, but are familiar with the path that takes them there. Our custom-built, highly scalable HD mapping system leverages machine learning, automation, teleassistance, and more to accelerate our ability to scale our business and give the Aurora Driver the resources it needs to move people and goods safely and predictably.

Virtual world-building: Revving the simulation engine

We run millions of simulations every day to train and evaluate the Aurora Driver’s software stack across a vast range of scenarios and driving conditions. It’s the quickest and safest way to find and address edge cases early, well before the software is loaded onto our vehicles. In this instance, our simulation experts walk through how we speed up development and validation through scalable simulation in virtual worlds.

Eyes on the future: Forecasting for human-like driving

Driving is full of complex interactions. The way you behave on the road is likely to affect the way vehicles around you behave. We’ve trained the Aurora Driver to predict how other vehicles will respond to its actions—and to each other—leading to more accurate forecasting and better, more natural decision-making. Take a look at this video for an in-depth explanation of our revolutionary interactive forecasting and learned decision-making architecture.

Feeding the beast: Processing data with advanced machine learning 

There's more to self-driving data than sheer quantity, as obtaining quality data enables us to accelerate the development of our autonomy capabilities and model performance. But when you’ve collected millions of miles of driving data, identifying which experiences are valuable can feel like finding a needle in a haystack. Our data teams have developed state-of-the-art machine learning tools that rapidly assess and process quality data, and that make the most of every valuable experience we capture on the road by turning a single encounter into the equivalent of thousands.

This is just a glimpse at how our teams are quickly developing our technology—frequently rolling out new capabilities that address edge cases, meet our partners’ needs, and more. We’ll continue to share updates as we progress toward the planned deployment of our driverless semi-trucks in late 2023 and ride-hailing passenger vehicles in 2024.

To see what else we shared at Aurora Illuminated, visit aurora.tech/ai.