Page 40 - EETimes Europe June 2021

40 EE|Times EUROPE

OPINION | EV & AV
cases. Field feedback on quantitative metrics
ADAS Versus Automated for performance will be essential. SPIs will be
fed back to the vehicle developers so that they
can understand how the performance assump-
Vehicle Safety tions change over the life of the equipment.
Another challenge is scalability past pilot
vehicles. Getting something to work for one
car or 10 cars, or even a hundred, is vastly
By Phil Koopman, Edge Case Research different from getting it to work for millions
of cars, where all the rare edge cases sud-
denly become daily events for some vehicle
THERE IS A DRAMATIC DIFFERENCE between safety require- in the fleet.
ments for cars with human drivers and those for cars that drive This will force a transition from brute-force
themselves. A look at the four modes of automated driving makes data collection to a more structured approach
this distinction clear. to talking about safety. There are going to
In assistive operating mode, the human drives the car and bears be common-cause failures across the sensor
the responsibility for safety. In supervised mode, the car does the modes, and you’ll have to think about those
driving, but the human’s eyes must still be on the road. An import- and account for them in your safety case.
ant line is crossed when entering automated mode: The human Some industry transparency is appropriate
“captains the ship,” but the car is responsible for both driving and to help the public accept that the right thing’s
driving safety. Finally, in autonomous mode, there is no human with any responsible role for the being done for safety despite the inevitable
operation of the vehicle. The car does all the driving, must ensure driving safety, and handles all business pressures. Safety collaboration,
other safety tasks. rather than competition, would be helpful.
There is tension between safety and permissiveness in automated driving, and the industry will When cars are fully autonomous, there will be
need time to resolve it. From a sensor perspective, this amounts to a tension between false detec- no human driver to clean up even the small
tions and false non-detections. False non-detections — false negatives, such as failing to detect a things that, in principle, can lead to a bad
pedestrian in the road — compromise safety. False detections — false positives — compromise per- crash if not mitigated. The automotive folks
missiveness. If you’re constantly seeing pedestrians that aren’t really there, you’re going to keep have to get it right. ■
stopping needlessly. In automated driving, you need exceedingly low false-negative rates because
you can’t afford to miss an object that’s important to avoid, but you can’t crank up your false posi- Phil Koopman is the co-founder of Edge
tives. Sensor fusion can help address the tradeoff, but it’s going to be a difficult one to get right. Case Research and an associate professor
Component makers must strike a positive trust balance. You need engineering rigor, valida- at Carnegie Mellon University. This pre-
tion, feedback, a safety culture, and a standards-driven approach to safety. Part of this is going sentation summary appears as part of the
to be safety-performance indicators (SPIs). Customers will need to know what they can expect proceedings of EE Times’ Roadmap to Next-
from a component in terms of safety, and they’ll want to see the data. Integrators will ask for Gen EV & AV virtual conference, now available
more than standards-conformance certificates; they’re going to want to know component safety for download on Power Electronics News.

IMAGE: SHUTTERSTOCK

JUNE 2021 | www.eetimes.eu