At Aurora, our mission and value proposition have been driven by safety since day one. Not only has this meant making safe autonomous vehicle operations the core of our business model, we’ve made safety the metric by which we measure the progress of our technology. As part of this commitment, we have pledged to only launch our driverless product when we can show its safety through the completion of our Safety Case.
A core part of this approach is rapidly advancing and validating the Aurora Driver’s capabilities in our industry-leading Virtual Testing Suite. Through simulation, we can both improve performance in the complex scenarios we have already seen on public roads and test capabilities against rare situations we haven’t yet encountered (or may never encounter!) in the real world. But where do these scenarios come from? And how do we know they’ll be relevant to our operations when we launch our product?
We’ve taken a multi-pronged approach to answering these questions, leveraging the expertise of safety experts at the National Highway Traffic Safety Administration (NHTSA) as well as real examples of fatal collisions between human drivers on our Dallas-Houston launch route.
Applying Regulator Expertise to Safety-Focused Development
Safety experts at NHTSA are not only federal regulators who oversee the autonomous vehicle industry, they are pioneers in helping the world understand and prevent vehicle crashes. As part of this work, NHTSA has developed a detailed list of crash types that reflect all law enforcement-reported collision scenarios that human-driven motor vehicles experience on the road. Aurora has made this data set a core part of our validation process – considering vehicle movement, actor placement, infrastructure, and more to develop autonomous driving technology that can potentially anticipate and, in many cases, prevent collisions.
NHTSA’s crash taxonomy details the collision types that vehicles encounter on the roadway – many of which are relevant to Aurora’s Dallas-Houston launch route. These graphics represent a few of these collision types.
To best use these invaluable insights, we are training the Aurora Driver to handle thousands of permutations of the NHTSA collision scenarios that are relevant to our Dallas-Houston launch route. We build these scenarios to be extremely challenging to help ensure the Aurora Driver performs reasonable maneuvers in all manner of situations. This is fundamental to understanding autonomous vehicles’ expected safe behavior in potential collision situations.
Many drivers may only encounter a roadway collision a few times in their life but, thanks to the expertise of NHTSA and our unparalleled autonomous vehicle development team, the Aurora Driver has learned from millions of these scenarios in simulation. This has led to the advancement of our technology and confidence in our ability to operate in the real world. Planning for how our autonomous trucks should perform in these situations is an essential part of how we complete our Safety Case for driverless operations on public roads.
Learning from Human Drivers’ Collisions between Dallas and Houston
While NHTSA’s crash taxonomy helps us anticipate what could happen on the road and prepare accordingly, it is also essential to understand what has happened on the road – providing the ability to learn from past collisions by human drivers and help prevent them in the future.
As part of our validation of the Aurora Driver’s expected performance at commercial launch, we examined fatal collisions involving human-driven Class 8 trucks on our Dallas-Houston launch route between 2018-2022. Using data available about these crashes, our team recreated the collisions in simulation to understand how the Aurora Driver would have acted in comparable circumstances. Based on our analysis, in the 29 instances in which the Aurora Driver could have been operating the initiating vehicle, the combination of its powerful sensors, attentive driving, and quick decision-making would have prevented the collisions.
Said simply, if the Aurora Driver had been in control, none of these fatal collisions would have occurred.
Based on a real fatal collision involving a non-Aurora, human-driven truck on the Dallas-Houston freight route, this simulation shows that the Aurora Driver would have been able to recognize the potential collision from long range and come to a halt before a crash with the truck occurred.
But we didn’t stop there. Just as we simulated thousands of permutations of the collision types identified by NHTSA, we created permutations of these Dallas-Houston collisions. By varying details, like vehicle speeds and positions, we build a deeper understanding of how the Aurora Driver handles challenging, potentially dangerous encounters. Throughout this testing, we have seen consistently strong performance that reinforces our confidence in the safety benefits of Aurora’s autonomous trucks.
Bringing Virtual Learnings to the Real World
Using simulation as part of development and validation is essential because it enables our autonomous trucks to experience circumstances that only happen incredibly rarely in the real world. But, because our fleet is autonomously driving thousands of miles per day on average for customers, the Aurora Driver has encountered and safely handled some of these situations in real life as well – further confirming performance initially validated in virtual testing.
In one case, a large truck attempting to merge into the Aurora Driver-powered truck’s lane began to move over too early, creating a scenario in which the Aurora truck was in danger of being sideswiped. Having practiced similar scenarios in simulation, the Aurora Driver-powered truck moved over within its lane to provide extra space for the merging vehicle. The result? No collision, and a safer roadway environment for everyone involved.
In another instance, the Aurora Driver correctly anticipated that another vehicle would run a red light at an intersection – heading straight for the path of our autonomous truck and potentially causing a collision. Identifying the danger here, the Aurora Driver delayed acceleration, waited as the vehicle ran the red light, and then safely continued through the intersection afterward.
Additional instances of the Aurora Driver safely reacting to potential collision scenarios can be found in our product capability showcase.
Delivering Safety at Scale
As the Aurora Driver is deployed on new routes and in new operating environments, learnings from safety experts at NHTSA, simulated collision scenarios, and real-world driving will all continue to be invaluable. In fact, we’ve already conducted virtual collision testing for our Fort Worth-El Paso route and seen similar positive safety outcomes.
Safety-focused autonomous vehicle technology can help address the epidemic of fatalities we see on roads today. There are approximately 5,000 fatalities attributed to large truck crashes each year on American roads, and that number has been rising over the last decade. We need all the tools at our disposal to save lives on our roads, and autonomous trucks can and should be part of the solution.