A Tesla mystery: Why didn't auto-braking stop these crashes?

[Thrift Store Scott]

ztirffritz wrote:
The media is largely clueless about what Tesla are doing and don't understand what's happening. Throw into the mix vehicle operators who are negligent and you have situations like this. There's absolutely no excuse for a vehicle operated by a human while on cruise control to crash into an emergency vehicle...except for human negligence. If I'm driving on the highway and I see flashing lights on the side of the road I slow down and give the police cruiser a wide berth. Clearly the vehicle operator wasn't paying attention.

No disagreement here at all.

Back to the "why" though: I notice this only seems to be happening to emergency vehicles while they're in use, all of which presumably would have their flashing lights on at the time.

Since police have started using the all-LED light bars on their cars, I've said the light bars are FAR too bright at night to the point of being blinding, making it all but impossible to see officers directing you around an accident or standing in the road near their car. Light bar manufacturers need to add a "Night Mode" to their products such as the LED displays on car radios have in which the LEDS dim to half or so of their daytime illumination when the vehicle's headlights are turned on.

Could it be that the over-bright LED emergency vehicle lights are as blinding to the forward-looking sensors of Tesla's collision avoidance system at night as they are to human eyes?

[ztirffritz]

space-time wrote:
[quote=ztirffritz]
The media is largely clueless about what Tesla are doing and don't understand what's happening. ...

Do you?

I work for one of their competitors so I probably know something about self-driving cars.
I currently own a Corvair.

I know enough to question the media when they say the vehicle was driving autonomously with Autopilot engaged that you need to question all the rest of the details of the story. Autopilot is essentially glorified cruise control / ADAS in it's current version. The difference is that Tesla are building a closed loop system that is supposedly iteratively improving as it gains experience. The ADAS on a GM/Ford/VW/etc car is the same on the day the car is crushed as it was when it rolled off of the assembly line. Tesla are pushing updates that improve the system, similar to how a driver with a Learner's Permit learns everyday. They are doing something that no one else has attempted. I won't say that I'm an expert on autonomous driving, but I know more than the average person on the street. I've watched quite a few companies and college efforts burn out when they hit a wall. It's a bigger problem than most anyone wants to admit, and it will be a messy process perfecting it. It will make mistakes. The question is, does it make fewer than people do? I remember when CMU had a red Hummer with LIDAR mounted all over it assessing it's surroundings and it struggled with DARPA's tests. We've made the sensors smaller, but generally everyone except Tesla are going down the same path as that Red Hummer from CMU. Ford's Blue Cruise couldn't navigate a bend on a highway. Waymo can only drive in Chandler AZ. Chevy's Super Cruise (CRUZ sp?) is limited to highways that have been surveyed. They all have their strengths, but none of them are trying to do what Tesla are doing.

[PeterW]

But Tesla is essentially running a (at best) beta test on public roads. It should err on the side of safety and it seems to in most circumstances but has a problem with emergency vehicles. Perhaps a pause is in order to let them work out a few more bugs before resuming its use on the street.

[Paul F.]

PeterW wrote:
But Tesla is essentially running a (at best) beta test on public roads. It should err on the side of safety and it seems to in most circumstances but has a problem with emergency vehicles. Perhaps a pause is in order to let them work out a few more bugs before resuming its use on the street.

First, show me that the Full Self Driving Beta is the system involved in these accidents. In most cases, they are not. It's the "driver assist" Autopilot.

[NewtonMP2100]

.....maybe like 'Ross and Rachel'.....and it was on a....brake.....???

[vision63]

Nothing is perfect. How many times has the system helped avert disaster?

[RAMd®d]

I've never owned a Corvair.

Not auto-braking is only one of the questions I have.

I'm curious as to how the Tesla came to collide with the police vehicle, when it was (presumably) off the road.

She pulled over to the side of the road.

Typically the police vehicle would stop behind the first, most often slightly to the left to afford some protection for the cop as he stands near the driver's door.

Did the Tesla strike the left rear of the car or was it directly behind it?

Obviously the driver wasn't paying sufficient, if any attention, or there wouldn't have been a crash.

Was the driver thinking 'The Tesla has auto-braking so I don't have to'?!

I have no sympathy for drivers who while using Tesla's Autopilot feature, crash their car.

They should have known better, probably did, but behaved stupidly anyway.

The same applies to auto-braking, regardless of what part emergency lighting may or may not play in the crashes.

It's great to know what might be allowing Teslas to crash in these investigated instances, but none of that would seem to relieve the drivers of their ultimate responsibility — driving the car.

Safely.

[ztirffritz]

PeterW wrote:
But Tesla is essentially running a (at best) beta test on public roads. It should err on the side of safety and it seems to in most circumstances but has a problem with emergency vehicles. Perhaps a pause is in order to let them work out a few more bugs before resuming its use on the street.

Out of the approximately 500k Teslas on the road with Autopilot hardware V2, only about 1000 have access to the beta FSD (Full Self Driving). To date, FSD has not been involved in a single accident that I'm aware of. About 2 weeks ago Tesla deployed a driver rating system to owners who have purchased FSD or subscribe to the feature. They can use it to quantitatively grade their driving skill and then Tesla will begin allowing more drivers into the beta program based on the score in the driving rating system. My theory is that they intend to use this to show that FSD is at least as safe as actual drivers with apples-to-apples comparison via the driver rating system.

[kj]

I don't think a lot of this tech would have to work that well to outperform your average person under average conditions. People are terrible drivers because they think they can do 7 things while they drive.

[Acer]

Clearly this is a situation where the auto-pilot should get it right 100% of the time. It's a parked emergency vehicle, for crying out loud. Broadside of a barn obvious.