Hacker News Reader

Comment by tschellenbach | Hacker News Reader

43k deaths per year due to human drivers. DMV should stop sticking their heads in the sand. https://www.iihs.org/topics/fatality-statistics/detail/state....

View on HN

I don't know why something like this gets trotted out in every self driving thread as if it's some sort of clever point. One thing being dangerous does not automatically make the shiny new alternative safer. General purpose self driving literally doesn't exist yet, and when it does it might be even less safe than human driving. Assuming that it will definitely be safer because of some highly constrained testing doesn't track, regardless of how interesting the tech is or how much money investors hope to make from it.

Because it will not get solved unless we're willing to invest time, effort and some risk into developing AI

How many of those deaths per year are human drivers driving the same routes at the same hours in the same weather as Cruise vehicles? Without any normalization this comparison seems moot.

And how many are DUIs. They should really be comparing accident rates for sober drivers.

Why? Allow enough humans to drive and inevitably some of them will start doing so drunk. It's in our collective nature.

It's illegal and we send people to jail for it. What is this.

Making something illegal doesn't stop some people from doing it, so it's something that should be considered when looking at the overall safety of human drivers. One benefit of robots driving is that they won't get drunk, get distracted because they had a stressful day, get drowsy but still try to make it home, speed because they're late for an appointment, etc...

I don't get it, someone's clearly willing to break the law if they are drunk driving, unless you're banning steering wheels I'm not sure how that's going to help.

Why would someone drive drunk if they can just let their car drive them home while they snooze in the back?

The same linked website (iihs.org) says between 30-60% of all fatal accidents involve alcohol.

I strongly agree that the safety comparisons aren't anywhere near the same ballpark. They are just too apples to oranges.

All that said, I don't think that means we can't evaluate whether they're safe in the limited contexts they're being deployed in. We just have to be aware those statistics are only as good as their context (weather, geography, terrain, traffic patterns/devices/laws, social norms, etc).

I'm initially most hopeful for constrained environments (city, low speeds, parking) and Waymo in particular, because of the house amount of investment and very thoughtful and cautious strategy in terms of deployment.

Yes, but I think the relevant metric is deaths per 100 million miles driven. Self-driving cars haven't even traveled 100 million miles.

There are already land based transport options that are two orders of magnitude safer than cars: buses and trains, with no AI necessary :)

BTW, that's 100 million miles driven by humans before somebody gets killed by a car (the stat includes peds as well as vehicle occupants).

Though maybe good to have a system where a human remote driver takes over in cases where it's confused?

Cruise has exactly that. In this case however, the car decided to "pull over" autonomously after the incident, before a remote driver could get involved, and with that pulled the struck pedestrian 20 feet along the ground and pinned them under the vehicle. If it instead alerted a remote driver immediately and waited for their decision before moving again, this would've likely ended much more lightly.

its not really helpful for split second judgement calls like this. If your pizzabot is stuck behind another pizzabot or can't figure out something, ok. But this happens before anyone can intervene. The switchover + time to comprehend and react remotely aren't going to change anything.

Some of the players in this space do this but it doesn't work because driving via streaming video over the internet is nowhere near as effective as a human driver. And is simply not possible in a fast moving emergency, only being helpful if the car gets stuck and won't move

The concept of "human remote driver" isn't feasible. Imagine driving a car using a cell phone network Zoom call. It's less safe than just stopping.

They have that, and these cars completely depend on it for fundamental parts of the stack. If they detect that their understanding of the scene doesn't match their map, they (generally) have remote operators clarify the best course of action for the car. This isn't a remote takeover - they just plot a path for it or select from a number of different options the car presents to them, or give it the all clear to proceed.

I worked in the AV space for years - not at Cruise - but during one of my first driverless rides in a Cruise car, we ran into a few blocks of SF where the power was out. About 3 street lights in a row weren't functioning. At each one the Cruise stopped completely for about 45 seconds before crossing - even if there was no traffic coming from it's left or right. That delay was so that a remote operator could clear it to proceed after it detected that there wasn't a functioning street light where it expected there to be one. After the first street light we were getting honked at and swerved around by cars behind us while it paused and waited for clarification.

Comparisons between the safety of robotaxis and human drivers are strange to make when 'robotaxis' still depend on remote operators for fundamental parts of the driving task

Reply on news.ycombinator.com