New safety standard governs self-driving cars

Automated Driving Systems are eventually expected to drastically reduce the number of accidents and fatal crashes caused by human drivers, but what will be the standard for whether robot vehicles drive safely?

Part of the answer is now available with a new IEEE Standards Association working group 2846 standard for AV safety being reviewed by governments around the world.

Governments and industry players will presumably be able to review the standard to identify “reasonable and foreseeable” assumptions for kinematic properties of vehicles and pedestrians in a variety of roadway scenarios.

For example, how should a self-driving vehicle be expected to react while driving at 40 mph alongside a sidewalk where a pedestrian is walking and suddenly turns toward traffic? Or what are the reasonable and foreseeable assumptions merging autonomous vehicles should expect from other AVs on a freeway?

Until the 2846 standard evolved, carmakers and governments were developing their own assumptions but now can work from an agreed-upon approach developed by 30 different companies including carmakers, chipmakers and mobility-as-a-service providers.  The working group members include Arm, Ford and Uber as well as Zoox and Waymo.  The biggest chip designers in the industry are also in the group: AMD, Nvidia, Qualcomm and Intel.

Jack Weast, chairman of the IEEE SA working group 2846, said discussions about the standard began at the outbreak of Covid in early 2020 when various AV industry players were hitting a critical turning point when discussing standards of safety.

 “We got into this trap about safety of AVs,” he said in an interview. “Some were saying the only acceptable level of safety is perfection. But if that’s so, AVs would never ship.  So, we started thinking about what it means to drive safely in a societally acceptable manner.”

In the example of one AV following another on a highway, how closely it follows at a certain speed will rely on a set of assumptions in the standard.  “What inspired us was the psychology of how humans drive,” Weast said. “We realize it works because we make assumptions…We started talking with other companies and identified the same need. We standardized a set of assumptions we can all share and that’s what brought us together.  Within a self-driving vehicle there are models and these models are where assumptions are used [such as} how close is the vehicle following and what is the maximum braking speed.”

The process to develop the standard over two years led to an honest conversation about safety. “We had to get rid of the idea of perfect safety,” he said. Perfect safety would mean AVs would only be allowed to drive at 5 mph or even slower. Or not at all.

“We have this opportunity to save millions of lives and with AVs there’s going to be accidents and there will still need to be crashes,” Weast said. If there is a crash of a robot car, police and society at large- will need to investigate and find out what happened, whether the vehicle was traveling at a safe speed and made the right assumptions about deceleration.

Weast, an Intel Fellow and CTO for the corporate strategy office, said the standard is technology neutral and was adopted by 30 or so members with one vote per organization. “This standard provides a minimum bar and baseline” that governments can use in developing safety rules suitable to their localities.

So far, the working group has presented the standard to the United Nations Economic Commission for Europe, the U.S. National Highway Traffic Safety Administration and officials in China, he said.  Various early academic papers and research were used to develop the standard, including the Responsibility-Sensitive Safety model for AVs proposed in 2017 by Mobileye.  A more recent paper from IEEE SA is available at no cost online.  

RELATED: Intel’s Mobileye tests AVs in ‘huge headache’ of NYC