Not sure what the author expects the right behavior would be here. Or what the point of the article is.
Author was riding on a motorcycle with the feature enabled, and dropped his phone in a way that was indistinguishable from a crash (which he admits). The author even says he's grateful the feature exists.
But then says "But surely there's a better way of detecting real crashes". Okay? What do he think that might be? No suggestions are made at all.
He does say 'Wouldn’t you think you’d get a notification on your Apple Watch, allowing you to dismiss the crash alert'.
Notification on the watch is a reasonable and safe change.
I wonder too whether a filter could be applied for situations where the phone detects massive acceleration, but the watch does not. It's hard to imagine a scenario where 1) a crash occurs, 2) the phone imu detects it, but 3) the watch imu do not.
Once the rider is thrown off the motorcycle, the motorcycle will eventually crash, imparting force onto the phone while it is still attached to the motorcycle.
If the phone is in the pocket of the rider at the time of accident, this can be detected using the proximity sensor or nominal accelerometer data crosschecked with GPS velocity, or even capacitive sensing.
EDIT: There are only three possible states before the crash. Either the phone is in the rider's pocket, or it is mounted to the vehicle, or it is held in the user's hand. If the phone is in the rider's pocket, it is going to stay there until the rider feels a strong impact. If the phone is in the rider's hand, then the phone should turn this off completely because it's far more likely to slip. If the phone is attached to the vehicle, it should only register a crash if it begins simultaneously with the phone being disconnected to the vehicle (if it is to begin with).
The fact that the three may be separated is irrelevant, because they will be separated after the crash begins. As long as the phone mount is sturdy enough not to separate under normal conditions, this is fine. If it is not, then a false positive is much more likely than a true positive. And a crash sensor with more false positives than true positives is actively harmful.
This only applies to two wheeled vehicles, obviously. For four wheeled vehicles, the way it's working right now is fine. It is pretty trivial for a phone to make the difference between a car and a bike.
It doesn't sound like you've ever seen a crash.
If the phone is on the pocket it can easily get separated from the ride before the rider feels a strong impact.
I'm not sure how you think the phone should figure out if it is disconnected from the bike.
While it may be possible to improve crash detection. I don't think any of your suggestions come close. And generally will make things worse
Why? Apple's implementation is clearly suboptimal. Motorcycle crash sensors aren't new, and they work differently to this for very good reason. Sensing whether the phone is on or off the motorcycle at the time of the crash is a really basic feature, because phones fall of bicycles and motorcycles almost as often as there are crashes to begin with.
A feature with that many false positives can be actively harmful. It's not responsible engineering to ship it. Especially if you lock it to specialized hardware, then you have no excuse not to go all the way and at least have a proximity sensor at the back.
Why would we assume the only possible states for the phone in a motorcycle crash would be attached to the motorcycle or attached to the rider? It's perfectly within the realm of possibility for the rider, cycle and phone to all be separated from each other.
And if its sitting in a cupholder in a car? If the phone is in loose in a saddle bag? In a backpack on the floor of a car? In a backpack on their back, out of range of a proximity detector? If the motorcyclist hits the phone on the way down and the phone goes tumbling and ends up hitting the ground in a way indistinguishable from the situation OP had?
Apple clearly chose to accept false positives in order to avoid false negatives. I'm pretty sure their engineers and lawyers have given this more thought than you have from your armchair in the one hour since this was posted.
The phone will receive a very strong acceleration before being thrown off the bike.
In this case, the phone is being disconnected from the bike while receiving a small amount of force, and then hitting the ground with great force.
EDIT:
In a one vehicle accident where the driver lowsides onto gravel, the phone will receive a lot of forces as the bike deccelerates or tumbles, and this force will start when the phone is still attached to the bike.
And how different will it be from what happened here? Will it be substantially different versus a one-vehicle accident where the driver spills on a gravel road?
The idea of a highly integrated ecosystem is all about providing elegant solutions to corner cases.
If you don’t have a cellular watch it’d just keep on working like it has been. False negative and all.
But if the watch was there… Crash detection could just use the “everything is fine” feature by pushing it down from the cloud to both phone and watch simultaneously.
> He does say 'Wouldn’t you think you’d get a notification on your Apple Watch, allowing you to dismiss the crash alert'.
As you lift up one hand to look at the watch, and the other to dismiss the alert, suddenly you are completely unable to control your motorcycle and then really have a crash!
You can stop the vehicle. Which (hypothetically) also serves to have you closer to the phone that may be findable and/or salvageable (if the phone can ring). Chances are not high, obviously.
If you go over 100 kmh, that means ~30 meters per second. Even with instant reaction time to the watch ringing, then you having a look at it and start breaking in a safe manner, you are already way out 100+ meters.
A lot of people will walk more than 100 meters to have an ice cream worth $3. A phone worth $1k+ (stupid), might be worth a stroll. That of course depends on the road and safety etc, I only mention this last point to save you using it as a empty retort.
That leads to a consensus problem. If you have two sensors reporting different circumstances, which one do you trust? Do you err on the side of assuming the crash report is a false positive? Then you may fail to report a real crash when you really have a false negative (the non-crash report). For the feature to be useful, with only a pair of sensors and no reason to assume that the one is giving a false positive, you should err on the side of caution.
Motorcyclist crashes, Apple watch wearing arm is severed but it lands in the bed of a truck and carries on down the road.
Imagining scenarios that would be workable, that's harder. The chances of the above are not zero... stranger things have happened (and that's from what we've seen on video, let alone what's not captured on film).
Why is it the author's problem to develop a better system? Is it not valuable to remind fellow users of the caveats of using the feature and provide examples of possible failure modes that might not be immediately obvious?
Why corporate apologism? We are talking about a trillion dollar company that just publicly unvealed a new feature that has shortcomings: do we need anymore justification to voice the shortcoming and hope that trillion dollar worth of resources may improve the feature?
[Edit] to clarify: trillion dollar company or not, a new feature was just released a someone noticed what seems to be an edge case where does feature isn't providing value. Feedback is good.
Vehicle crash detection devices existed for years. Yes, you have to install it, but you can't forget them at home, don't send an alarm if your phone falls and can call emergency services automatically, so nothing new.
There is definitely a better way to do so. You can use a magent with a force sensor to detect when the phone is against a metal surface, and only trigger the crash detection if there is sizeable force before the phone detaches from the vehicle.
Then Apple would actually be justified in locking this feature to some models, as if normal accelerometers didn't already go to 16G+ :)
EDIT:
I feel like I may have been unclear. I'm saying the phone should detect if it's being detached from the bike before or after the crash, using a magnetic sensor preferably (since Apple already has their magnetic attachement thing).
If the phone is disconnected from the vehicle before it senses a crash, then it's a false positive. If the phone is disconnected at the same time, or less than a certain before sensing a very strong shock, then this is a true positive.
Here's the fix: Secure your hardware. It's not Apple's problem if you do a poor job. He even stated his belief that the adhesive wouldn't be up to the task.. and he was right.
"I know false-alarms like mine will be rare, and the potential life-saving benefit far outweighs the inconvenience of my particular situation. But there has to be a better way to distinguish genuine crashes from phone-smashing mishaps."
Just want to stress that training an ML model is going to be really difficult for this scenario, and false positives are to be expected. It isn't like this was unexpected either given that "the phone flew off its handlebar mount".
The most amazing thing about the story is that the iPhone was ripped from its adhesive mount by a bump, flew off the motorcycle, presumable struck the concrete road at 60 miles per hour, and still had enough functionality to activate the emergency crash detection and find and text contacts marked as family!
Honestly, what is the best case scenario for this feature? Sure, text the police, that’s a great idea, but why give your family extreme anxiety with very little information? What are they supposed to do, rush to the scene of the accident? There’s nothing you can do when you get that text but wonder if they’ve died.
You can text them immediately to know you are fine. Nonetheless, this sort of false sensitive is very unlikely with cars, and chances are you overestimate your well-being in a car crash due to adrenaline, so even then it may be good to have your family look after you more than they would otherwise.
This is a good example of an unanticipated or rare scenario that nobody would have realistically considered when developing and adding a feature. The author's suggestion of displaying an alert on an Apple Watch to seek confirmation is a good idea, and will certainly appeal to Apple sales philosophy of integrating services to sell other device. :)
B Franklin said that someone who is willing to give up freedom for security deserves neither. This rider didn't get the security for which he surrendered his freedom and got worse.
Don't use this feature with a handlebar mount, I think is the lesson. I suspect 99% of these reports are going to be dropped phones and police will learn to ignore them.
It isn't possible for crash detection to differentiate between "phone falls off bike" and "bike crashed" because everything that happens to the phone could be identical in both scenarios.
If the phone is loose in a pocket (or even in a backpack) when the bike crashes, it can easily hit the ground, bounce, and then come to a stop... just like it does if it falls out off the bike in any non-crash scenario.
I keep procrastinating putting a mount on my bike, so now every time I go out for a drive I need to pull over, take my phone out of my pocket, check to make sure I didn't miss the turn (50/50 chance), put it back in, and get going again - only to repeat the process whenever I feel lost again.
Author was riding on a motorcycle with the feature enabled, and dropped his phone in a way that was indistinguishable from a crash (which he admits). The author even says he's grateful the feature exists.
But then says "But surely there's a better way of detecting real crashes". Okay? What do he think that might be? No suggestions are made at all.