Frmr AP Engineer on Brown Crash: Car thought Truck was a bridge - reason AP brakes for overpasses?

[snakeopus121]

When I read this - immediately thought this may be the reason why AP brakes for overpasses:

"With the Brown incident, Kouri said, his Model S thought the underside of an oncoming truck was a bridge. If it had access to high-definition maps—maps that are constantly updated—the car would’ve realized there was indeed no bridge and hit the brakes, Kouri believes." ---Andrew Kouri, former AP engineer.

http://jalopnik.com/meet-the-former-tesla-engineers-who-want-to-fix-self-dr-1797059589

 

[animorph]

Absolutely. You want the car to brake for obstacles. But the flip side is that you don't want it to brake if there is no obstacle. Kind of like filtering for spam email. AP is not 100% sure in some cases, so you may get the occasional wrong decision.

Back when this was first implemented, with radar-only braking allowed, there was going to be a white list of known false alarm locations or radar false alarm signatures. That would allow the car to ignore known false alarms. That's one way to reduce the braking false alarms.

This is a hard problem for the radar primarily because it can easily differentiate objects by their relative speeds. Unfortunately there is a ton of stuff moving at the same speed as the road surface, including signs and stopped cars. It is not so good at differentiating between those signs and stopped cars. Tesla is still working on it.

There is again a flip side to the false braking problem: will the car stop for an obstacle in the road if it needs to? More false braking, or a greater chance of a collision? The answer of course is to improve the obstacle detection to minimize the chance of either problem, but that may take a long time, or more hardware.

 

[Tam]

...If it had access to high-definition maps...

His solution for inferior sensors is to supplement them with a map.

From a safety view point, I would prefer to get better sensors, like more cameras for different views, more radars for different angles, and of course LIDAR.

Once the capacity limits of sensors exhausted, then I would go for maps as the last resort.