Waymo claims to already have a better record than human drivers. How long before we read a similar case where a driver is accused of creating unsafe conditions by taking control of the car?
> There must be a vehicle in front of your car, reasonable road conditions with readable markings and lines, and clear weather and light conditions. Drive Pilot can’t be used at night or in the rain, and the headlights and wipers must be set to auto for it to work.
> It’s also only available on freeways that have been mapped by Mercedes, with GPS positioning that is precise to the centimeter and even accounts for continental drift. Drive Pilot can’t be used in construction zones.
Mercedes is actually being cautious about slowly rolling things out, not just yeeting it over the fence like Tesla.
Little things like:
> Mercedes shows us an EQS with an important visual indicator that is not yet legal: in the hopefully very near future, when Drive Pilot is active, the headlights, taillights, and side mirrors will have turquoise marking lights so other drivers know the Level 3 system is in use.
and
> Drive Pilot also has redundant braking and steering actuators and a separate onboard electrical system just in case one of them fails.
are good things.
And it drives better than Autopilot, despite being "glorified cruise control":
> It does feel more precise and accurate in how the car stays in its lane and reacts to surrounding traffic, with fewer jerky movements and constant tiny adjustments. Drive Pilot even reacts to larger trucks or motorcycles that are lane splitting by moving slightly over in the lane without crossing the lines, and not only does it work in carpool lanes but also it can tell the difference between the carpool and FasTrak lanes and will let the driver decide which to use.
This doesn't sound like Tesla at all:
> and there’s not a single moment where Drive Pilot makes a mistake or reacts badly to a situation, even when surrounded by nightmarish LA drivers.
Some of us applaud these kind of things.
Remember the whole phantom braking thing a while back? And then it disappeared with an update, but another phantom braking problem arose, so then they reverted that update, and then released another update that fixed those two problems...
A safety culture where a component manufacturer had firmware that took around 36 hours to work through the full test suite... released a new firmware to Tesla in response to issues. Got an email from Tesla 3 or 4 hours later saying "Thanks, this is working so much better", and when they responded with confusion about the test suite, etc., "Oh, we just flashed it onto one of our cars here and took it out on the road and tried".
They are the only ones using the consequences of general principles of product liability as a selling point, might be more accurate.
I'm worried about a time when the _statistical_ safety of AI driving is better than humans across the board, but there are horrifying correlations between AI induced accidents, like "Children on tricycles are not well recognized" or "slight fog in Seattle induces 150 car pileup". Or, AI causes worse accidents less frequently than humans. Whereas humans have lots more fender benders and minor injuries to older adults.
Not saying this is true, but it's a scenario I think about a lot.
Continuing your example just for exposition / hypothetical, if those are 100 individual deaths with varying causes, vs 99 children run over by AI b/c they were walking in a line holding hands and were interpreted as road markers, well that's not _better_ is it?
This is purely hypothetical, but statistics gloss over those human details.
To me, more sensors means faster implementation. Quicker to get on roads. Less total deaths. At least in theory, and in what successes i'm seeing.
If it get's cheap enough it will be.
But the AI isn't human and things a human would recognize, like a a large grey truck up ahead doing a turn on an overcast day an AI can just miss.
> if you say we can't have automation until its 100% perfect,
That's why we need the lidar.
99.99% of daytime highway dashcam frames don't include a firetruck stopped in the lane, dealing with an accident.
But let's say, hypothetically, that a self-driving car accelerated into a clearly visible, stationary firetruck in clear daylight. Despite being 99.99% perfect.
Legislators might feel in order to be an above-average driver the vehicle shouldn't hit any clearly visible stationary objects in clear daylight, because the median human driver doesn't do that.
Elon Musk claims the same but there is a "terms and conditions apply" and they don't agree to assume responsibility.