Troubles with Tesla's automated parking summon safety regulators
reuters.com
reuters.com
The warnings say you need direct line of sight, that the system isn't perfect, and that it may not detect all obstacles. Even ones expected to be in parking lots. Those warnings all make sense, but the people recording the videos don't care. They're just pressing the button and being shocked that it doesn't work.
Ultimately, regulators will step in if they feel that people still get into crashes despite having the warnings. If the warnings don't stop people from doing stupid things, they'll require more warnings or kill the feature. Unfortunately for Tesla, the regulator focus is public safety (and the underlying statistics). If the tables turn then it doesn't matter how safe the feature is when used correctly. Instead, it matters how safe the feature is when used incorrectly.
And that will only hurt Tesla in the long run. And that's a shame, because it will make true self driving cars that much harder to get to market.
You cannot blame people who bought a car because it has a feature they want actually trying to use that feature.
No, they will happily accept the third option: a product that doesn’t require warnings.
Cars must have seat belts, ABS, airbags, crumple zones, etc. You can’t buy a chainsaw without kickback reduction, heavy presses are specifically designed so that you need two hands to operate them, etc. etc. There are zillions of cases where we deem warnings insufficient.
Why would this feature, that even can injure bystanders, be different? It’s not as if its benefits are so large that they obviously outweigh its risks.
It always used to be the case that OTA-updateable cars are one security incident away from becoming a car botnet, but now that "summon" exists, it's theoretically possible to hack a Tesla to hunt humans by camera and other sensors.
“The Pedestrian Warning System (PWS) is an audible tone played when the vehicle is moving slowly (up to 19 mph) in either Drive or Reverse. The sound is played via a speaker mounted in the front fascia and alerts nearby pedestrians of the car’s presence in low-speed situations.”
Easy to unplug, but I don't recommend doing so depending on your state's liability laws. The pedestrian takes on a percentage of liability when they're in the roadway in several states.
https://electrek.co/2019/09/10/tesla-pedestrian-warning-syst...
I am a cyclist and a pedestrian, and I hate this entitled attitude that a minor unpleasantness is worth more than my safety. Even with the noise, the cars are much much quieter than an ICE car.
They’ve been doing this with Autopilot since the beginning: they have the demonstrated ability to restrict Autopilot to highways, which is the only intended use, but they choose not to flip this switch, so that they can use customers (and you, me, and your children) as guinea pigs and collect that sweet, sweet data, which they upload continuously to their servers.
They certainly have the front seat compared to any other competitor when it comes to real world scenarios.
IMO if Tesla were really confident in their summon technology, they would cover accidents with it themselves. Relying on errors with their technology to be covered by outside insurance is reckless.
This new feature though is not available in Europe and I guess it won't be until the European manufacturers start having the same features :(
This occurred to me as well. I have wondered if Tesla started their own insurance pool as a contingency plan to address the risk that other insurers will begin to decline coverage or price it impractically, or to show that it's possible with reasonable prices.
Or exclude use of FSD from coverage.
I wouldn’t call it “FSD” except as a joke. Tesla is 15 years away from FSD.
Mathematical certainty, or even criminal-style proof beyond a reasonable doubt isn't required. The insurance company writes the rule in their policy, refuses to pay if they think there is sufficient evidence that the policy allows it, and if the insured disagrees they get to sue the insurance company over it, and the case will be decided on the civil preponderance of the evidence standard.
Your car is not a fucking toy. He could have injured the occupants of the other car while he watched and filmed because he wanted to play with a cool feature without first learning how it works.
If you don't know what's going to happen, don't assume your automated car knows how to avoid a collision.
Tesla on the other hand can freely release half baked software killing people, causing accidents with being 100% at fault and now threatening kids in shopping malls without any consequences.
I love what Tesla is doing to the auto industry but I don’t like the idea if a company that’s not accountable.
1. Legally, just because you're a licensed driver does not mean you've been licensed to remote-operate a car, while standing on the sideline, in a pedestrian-heavy environment, where your line of sight, and your vehicle's path of travel is obstructed by hazards.
2. Teslas seem incapable of reliable collision avoidance, so you can't just expect your car to do the right thing.
Was the human at fault in the video I saw? Yes IMO. But even the Tesla owner who shot the video posted it saying he wasn’t sure whether his insurance company would consider him or the other driver at fault.
https://twitter.com/eiddor/status/1177749574976462848
The driver had to hit an e-stop, to prevent an accident.
And also, while we're at it, please explain how a self-driving car that is incapable of seeing that it's going to get t-boned, if it crosses an intersection where it does not have the right of way is in any way ready for the public.
https://www.freep.com/story/money/cars/chrysler/2016/06/20/j...
Laws certainly apply, but not necessarily "normal laws governing motor vehicles". For example, you probably can't be ticketed for running a stop sign in a parking lot on private property. However, you would almost certainly be liable for damage if running that stop sign caused an accident and you could be potentially charged with some type of generic reckless driving offense. Obviously the laws vary by jurisdiction.
Here (Australia) the rules define "road related areas" as including private property that does not have a "normally locked gate" protecting access.
This means all road rules (including stop signs, alcohol/drug rules, mobile phone use, and speed limits) apply in places like publicly accessible carparks on private property, and homeowner's driveways if they do not have a "normally locked gate". People have been booked for drink driving and mobile phone use sitting in their cars in their own driveways here (almost certainly after "failing the attitude test" and pissing a cop off enough for them to punitively enforce a stupid interpretation of a poorly written law, but that's a different rant...)
And law enforcement investigates and assigns fault in reports for accidents on private property even where vehicle code (other than hit and run and DUI, which IIRC apply everywhere) does not apply.
It's probably not clear yet how safe the "summon" feature is. Presumably the sample of it's performance is still small, but those videos don't look very encouraging, not just in the sense the car made "mistakes" like failing to yield, but also qualitatively. I would certainly have misgivings about being responsible for the results.
It seems like a less risky strategy to improve fully autonomous operation this way then on the freeway, but still risky. Perhaps the thinking is that parking lots are expected to be low speed, low stakes environments, and the car is pretty good at detecting pedestrians and avoiding low speed collisions, so the likeliest problem is property damage. However, people have a way of occasionally doing things that are hard to predict or imagine when we are defining our safety assumptions.
In the worst case, a wrongful death suit or something similar might be affordable in dollars but probably not publicity. Even a few less severe accidents seem likely to test the patience of both regulators and the public. Maybe autopilot will get better with more real world use and training data. Or maybe it won’t, or not fast enough. I’m really curious to see how this turns out, but not be being personally responsible when somebody gets hurt :(
It is one thing to test "beta" software on your PC but it is a totally another thing when it can actually kill people.
They should look at how medical devices are developed and certified.
Does anyone really think that will become a reality before the cars end of life?