821–830 of 1001 posts
From my perspective as a person who lives in San Francisco and also drives a LOT (10-20k miles per year, and many small drives within the city): Cruise cars do not perform acceptably. They manage to avoid collisions by driving extremely conservatively, but the way they traverse, say, a left turn against traffic is absurd. They slow everyone down, including emergency vehicles and public transit, by performing far belo…
What I found so far is this:
- Waymo and Cruise have driven about 8 million miles driverless (mostly in San Francisco) and reported about 100 crashes (mostly minor like scraping shopping cart, etc) and many that were fault of another car. (as of September 2023)
- Apparently a human drives am average 100 million miles between fatal accidents. In that case we do not have enough data, yet.
I'm not sure what to think. Is the rate of accidents a cause of concern? Or is this blown out of proportion and driverless cars are actually safer? Or (most likely) we simply do not know, yet?
From my perspective as a person who lives in San Francisco and also drives a LOT (10-20k miles per year, and many small drives within the city): Cruise cars do not perform acceptably. They manage to avoid collisions by driving extremely conservatively, but the way they traverse, say, a left turn against traffic is absurd. They slow everyone down, including emergency vehicles and public transit, by performing far belo…
From my perspective, as a relatively early adopter of Waymo (60+ rides). I have zero gripes with the driving itself. In fact I've seen Waymos do things that no human would be able to do*. Drop-offs and pickups I'd like to see improvements made like getting a bit closer to the curb. Occasionally the route selections make no sense at all. * Waymo made a right turn up a steep hill. Two lanes each way. It then pulled a b…
Wonder what it would have done if it didnt have room to pull into the left lane without hitting a car. Or perhaps another skateboarder.
Earlier quoted context omitted.
I saw a Waymo come to a screeching halt in a split second as a... less than intelligent individual... skateboarded in front of a bus, into the extremely busy road . If it was a human that person would have been dead, no question. Sometimes I'm riding in a Waymo (which I do every single day, 4x) and it does something, and I double take. Then a second later I see whatever it was it reacted to, and I'm like "dang, why d…
Waymo is amazing, knowing some of the stuff they do behind the scenes to ensure safety - I would feel safer riding in a Waymo than driving myself. My biggest fear has always been that Cruise or Tesla would shit the bed so bad we don't get any self-driving, either because of regulatory constraints or ruining the public perception of them.
Such as?
I'm disappointed to hear this news. So many people are killed by cars every year. We just take that for granted. Self-driving cars have the potential to drastically improve this and save many lives. I worry that these regulators are weighing the costs heavily while ignoring the benefits of innovation. That said, I don't have access to the data and I don't really know if Cruise has been acting inappropriately. I'm gla…
I'm not honestly. Cruise was pushing too hard and fast for where their stack is at and they seem to have some severe safety related bugs. Ex [1]: The car randomly starts swerving rapidly across traffic lanes towards an occupied sidewalk. Ex [2] (@1:31): The car doesn't stop for someone already in the crosswalk and nearly hits them. [1]: https://www.dailymail.co.uk/news/article-12640657/self-drivi... [2]: https://abcn…
For some reason, I can't imagine literally any human driving quite like that?
Earlier quoted context omitted.
> Another fun anecdote. When a Waymo is in the vicinity of another Waymo they acknowledge each other in a whimsical/cute way - the windshield wipers swipe once (I'm not sure they still do this). What is your motivation here? I can see why some think it looks like promotional material.
What do you even mean by motivation? If it's true (though it's probably a sensor error), it's perfectly reasonable to mention. And lying about it would be ridiculous. What scenario is there where you're worried about bad motivation here? It's not unrealistic for someone to like the idea.
Earlier quoted context omitted.
If feel like Seattle needs an all day and weekend Sounder service, more than it needs self driving robo-taxis. 1.5 hours away you might be at the edge or beyond Sounder reach so this may not help you much unless you get a local bus to your closest station, but you can probably still just drive to your closest station and not having to have to deal with driving in the city. If the Sounder would run like the ferries do…
I am staring aghast with European eyes at a city of 4 million people where the commuter rail doesn't run outside peak hours. How do you get home if you go out for a couple of drinks with colleagues, or stay late for a call? You shouldn't be asking for a couple of trains between 8pm and midnight. They should be every half hour, or more! Midnight until 6am on Saturday and Sunday could be every hour or two. (I live in C…
The Seattle area has a newer, dedicated light rail system (Link) that runs every 5-10 minutes from about 5AM to about 1:00AM. The stations are located much more conveniently and frequently, because they were placed based on where a commuter train should stop, not trying to piggyback on existing freight lines. There still isn't as much parking as there should be, but at least the stations are placed so that more than a tiny handful of people can get there without driving and parking.
Earlier quoted context omitted.
Perhaps we could optimise it further, by having sets of steel paths for the cars to drive on too.
Or we could link them together to make it safer so one doesn't veer off the track, maybe get nicely furnished stations where people can switch over to another set of linked cars to get where they need to go. Imagine the possibilities!
Earlier quoted context omitted.
It's worth noting that both Waymo and Cruise do report a shit load of statistics to regulatory bodies, including all incident reports. Neither company has caused a fatality to my knowledge.
Waymo did run over a dog https://techcrunch.com/2023/06/06/a-waymo-self-driving-car-k...
My drive to work is 20 miles and a really good stretch of that is just 1 lane where I’m stuck behind a car or the road is empty. In those cases I have no problem letting FSD handle everything, the road is extremely hilly, bendy, lane lines vanish every once in a while and it’s fine so I do find it useful at times, though not worth 200$ a month by a long stretch. I’ll probably cancel the subscription soon.
Its problems appear very clearly whenever it needs to accurately predict what other cars on the road will do to make its decision. It is honestly horrible at that and while a few times I tried to not intervene, I realized I was just creating a nuisance and now stopped allowing FSD to deal with those situations. It is also bad if there are wrong lane lines (happens at one final stretch of my daily drive)
Finally, as someone working in AI, I’m actually very optimistic about their next approach of just learning an end to end model to drive the car. I think this is the only scaleable way to reach human level driving. Any set of rules based + optimization based planner will have those edge cases that ruin the cars capability, in fact it turns out the long tail is really long. When a human is not drunk and focused he is a remarkably good driver. On the other hand, if you just need to capture the edge cases data and add it to your training set to solve it, you can capture all the possible driving conditions that come up on road. This reminds of how there were hundreds of failure modes and jailbreaks for ChatGPT early on while now there are close to None, presumably because they just added them to the training set. If a good human level self driving car appears, I’m convinced it will be an end to end approach, whether the Tesla team are capable of this, who knows.
Also for those who think, OpenAI’s scale based approach only works for language, I have access to ChatGPT Vision and its capabilities astound me. Scale for sure works