Re:Tesla's 'self-driving' software fails at train crossings
by quantified on Wednesday September 17, 2025 @02:03PM
(#45279209)
Still, since "full self-driving" is allowed to mean "not at all full self-driving" by our laws and courts, the driver is responsible for any accidents here.
Re:Tesla's 'self-driving' software fails at train crossings
by BinaryIgor on Wednesday September 17, 2025 @01:56PM
(#45279131)
I sometimes wonder whether we will ever solve all edge cases like the one described here; maybe, autonomous driving will always remain semi and almost there, but never quite fully
Re:Tesla's 'self-driving' software fails at train crossings
by yorwba on Wednesday September 17, 2025 @03:03PM
(#45280002)
Waymo uses remote operators to resolve situations that the car can't handle on its own (maybe that includes train crossings), so it doesn't need to be fully self-driving: https://waymo.com/blog/2024/05/fleet-response/
Re:Tesla's 'self-driving' software fails at train crossings
by Zigurd on Wednesday September 17, 2025 @03:32PM
(#45280392)
Nobody hasn't got remote operators. The difference is whether there are a few enough of them for the fleet that scaling service areas doesn't break the bank.
Re:Tesla's 'self-driving' software fails at train crossings
by Zigurd on Wednesday September 17, 2025 @05:45PM
(#45281763)
FSD for private cars requires you to be the monitor. Not remotely. But sitting right there in the driver's seat with your eyes being monitored for attention.
Tesla Robo taxis have remote monitors and have a safety driver, either in the passenger seat or in the driver's seat.
I've seen it suggested that Tesla remotely monitors demo rides. But they don't officially say so.
Re:Tesla's 'self-driving' software fails at train crossings
by rsynnott on Thursday September 18, 2025 @09:35AM
(#45289521)
This doesn't seem like that much of an edge-case. Level crossings are not exactly some weird exotic rarely-seen thing, and you'd think they'd actually be relatively easy compared to normal traffic junctions; they're clearly marked on the maps, and they have big obvious flashing red lights.
Re:Tesla's 'self-driving' software fails at train crossings
by duxup on Wednesday September 17, 2025 @01:58PM
(#45279152)
I'm sure it's more complicated than it seems (or maybe not), but that video is pretty damning. It's not like something is blocking the car's view, the lights and crossing bars are very visible. Car just doesn't notice / keeps going.
Re:Tesla's 'self-driving' software fails at train crossings
by jerlam on Wednesday September 17, 2025 @03:06PM
(#45280029)
I'm trying to understand if there is a situation where driving into a red and white stripe with flashing lights on it is correct. Does the Tesla think that because there isn't anything below the crossing arm, that the arm is actually farther away? Or it looks vaguely similar to the back of a trailer, but not enough so that is ignored?
Re:Tesla's 'self-driving' software fails at train crossings
by Doxin on Thursday September 18, 2025 @08:49AM
(#45288996)
As far as I can tell from previous failures, teslas will only attempt to avoid obstacles they recognize. How anyone could set up a self driving system like that and think it's a good idea is beyond me.
Re:Tesla's 'self-driving' software fails at train crossings
by FireBeyond on Wednesday September 17, 2025 @02:11PM
(#45279305)
Doesn't surprise me, yes, things have evolved, but 18 months ago in Pennsylvania, watching Tesla show a train crossing as an erratic conga line of trucks and an equally erratic looking traffic light.
Re:Tesla's 'self-driving' software fails at train crossings
by nrds on Wednesday September 17, 2025 @06:03PM
(#45281925)
You're talking about the visualization app. This has almost nothing to do with FSD besides being fed by the same cameras and maybe some similar preprocessing.
Re:Tesla's 'self-driving' software fails at train crossings
by FireBeyond on Thursday September 18, 2025 @10:06PM
(#45297225)
So you're telling me that the way it depicts traffic signals on screen, and the way it reacts to traffic signals, or alerts you, are completely independent?
Yeah, I fail to see any issue there. Like the car attempting to drive through the boomgate...
Re:Tesla's 'self-driving' software fails at train crossings
by SergeyKa on Wednesday September 17, 2025 @07:50PM
(#45282944)
Elon: buys some TSLA
TSLA: wheeeee!!!!
Shorts: oh sh!t, damage control now!
Opinion piece writer: here, thx for the dough!
NBCs and Reuters: posted, thx for the dough!
HNs and Reddits: thx for engagement!
Nothing newsworthy has actually happened, we're just feeding the trolls in shorts, and sending the clicks to justify the dough.
And Nashville: https://news.ycombinator.com/item?id=45275415
Waymo actually works, and is likely to be at or near breakeven financially.
The SFO permit also involves an initial phase using human safety drivers in the car: https://www.sf.gov/news-mayor-lurie-takes-key-step-toward-la...
Waymo actually works because they don't insist on automating what they cannot automate safely yet.
Tesla Robo taxis have remote monitors and have a safety driver, either in the passenger seat or in the driver's seat.
I've seen it suggested that Tesla remotely monitors demo rides. But they don't officially say so.
It's very strange. Even if it doesn't see it as a RR crossing, there's arms right there.
Yeah, I fail to see any issue there. Like the car attempting to drive through the boomgate...
Nothing newsworthy has actually happened, we're just feeding the trolls in shorts, and sending the clicks to justify the dough.