Three years ago, Intel and Mobileye introduced the responsibility-sensitive safety, or RSS, model to encourage the automated-vehicle industry to collaborate and align on what it means for an AV to drive safely. For an industry that normally competes on safety, the suggestion was met with mixed response. Some players readily agreed, while others felt strongly that safety assurances were brand attributes and therefore not open for review.
Meanwhile, multiple studies have shown that the public is wary of self-driving vehicles. The idea that a “robot” car could kill somebody is a frightening proposition. This is true even though we all know that human drivers kill thousands of people every day worldwide.
We also know AVs can — and will — be so much better drivers than humans. But how much better? How safe is safe enough before society will allow AVs to bring their life-saving promise to our roads?
This is a critically important question. And we believe the only way to do this is in collaboration with the entire AV industry and governments around the world. Continuing to work on safety in closed groups or individually won’t get us anywhere.
Many have said that “driving safely” is a risk balance between safety and the usefulness of a vehicle. We all make assertive maneuvers to get where we need to go.
Will society allow AVs to drive the same way humans do — assertive maneuvers and all — to get their passengers to their destinations? Or will AVs be required to adhere to more conservative rules, thereby keeping the AV from asserting itself in traffic, moving more slowly than other vehicles and hindering traffic flow?
Safety models such as RSS can help the AV achieve this risk balance. But the safety model itself is only part of the equation. AVs must make assumptions about the reasonable and foreseeable behavior to expect from other road users. Those assumptions — quantified in the form of performance parameters — can then be plugged into the AV’s driving policy via its safety model.
A forthcoming standard from the Institute for Electrical and Electronics Engineers — IEEE 2846 — will give regulators clear guidance on the assumptions they can use to decide how safe is safe enough.
The beauty of IEEE 2846 is that it is being developed in the open by representatives from across the global automotive and automated-driving industry, providing the necessary transparency and peer-reviewed confirmation that gives governing bodies assurance of broad industry consensus needed to set regulations.
This is the collaboration we’ve been advocating since we published RSS, and we are delighted to have more than 27 entities working together to solve this crucial challenge, including co-leaders Waymo and Uber.
Other standards efforts that are limited in membership to only certain kinds of companies or operating in service of proprietary solutions only contribute to skepticism and regulatory delays.
AV safety is a problem we all need to solve. Companies with a stake in the game should challenge themselves to transparently demonstrate in real-world settings how their safety models can not only ensure safety but enable naturalistic real-world driving. Even better, they should join in open standards such as IEEE 2846 to ensure that the performance criteria ultimately adopted into regulation is based on sound scientific methods, peer reviewed and proven both on paper and in the real world.
Time is of the essence. Multiple companies have started operating driverless vehicle services even though consumer trust in them is tenuous at best. One tragic accident may be all that’s needed to keep this promising technology off our roads. Are we really willing to take that risk?