e.w. niedermeyer: "It marks the second incident in two months involving a Tesla’s near miss-with a train while utilising its driver assistance system." You can tell Tesla is close to "solving self-driving" because they are down to the tough edge cases like *checks notes* trains nz.news.yahoo.com/tesla-autopi...

"It marks the second incident in two months involving a Tesla’s near miss-with a train while utilising its driver assistance system." You can tell Tesla is close to "solving self-driving" because they are down to the tough edge cases like *checks notes* trains nz.news.yahoo.com/tesla-autopi...

Tesla autopilot appears to veer electric vehicle onto train track it mistook for roadnz.news.yahoo.com Local police urged drivers to remain ‘vigilant while using Tesla's autopilot feature,’ noting that it ‘can fail’

Earlier this year I was supposed to go watch the trial brought by the parents of Walter Huang, who died in 2018 when his Tesla on Autopilot thought the "gore" between two lanes was a lane and slammed him into a barrier. Tesla settled that rather than further reveal how crude these systems are.

You are trusting your life to a system that thinks "well, two lines roughly that distance apart on what is not explicitly not driveable space is probabilistically a lane, so let's just lock onto that and see where it takes us" and yeah sometimes those two lines are fucking railroad tracks.

The bullshit part of this is that Tesla absolutely buys map data as an input for these systems, so the "excuse" (it isn't one!) that they are just doing everything with computer vision isn't even valid. If you can buy traffic data you can goddamn well buy map data to validate the ODD, scum!

Exactly!!! Exactly!!! My phone Can tell what road I’m on!!

Until quite recently, my phone would tell me to take a "side street" off the FDR parkway on Manhattan's East Side, to get to the Holland Tunnel. The "side street" was a fish market lane some 50 below the elevated parkway. If a phone can't get it right, I'll never trust one of Elmo's death mobiles.

In my small town in France, Google kept sending me to drive through a small medieval foot lane / stairs that in addition is blocked by renovation

I used to drive for Uber and door dash and the nav would try to get me to do some seriously goober shit. With all the info they steal and all the input of people trying to correct maps, they still fuck up. We’re so far from being to safely have self-driving cars.

In my experience, these navigators are terrible with Chicago's lower Wacker drive because the GPS gets hella confused about which vertical road layer you're on

not a lot of traffic on train tracks, so it's a very effective shortcut. brilliant gambit, sir

"ah probably one of those commonly seen traffic lanes that's exactly 4' 8.5" wide, that makes sense"

The really cool thing about re-architecting the entire system to be "end to end" is that there is literally no way to simply go into guts and specify critical distinctions like "this is a road but that is a railroad track" in a discrete way. You literally have to reroll the entire stack.

Do they still judge the distance to stop size by apparent size?

I don't believe self driving code is impossible. I do believe it is impossible under Elon Musk's "management". Picture him firing any developer who won't promise to meet his idiot deadline. Keeping those who will. It's the same at Boeing. Only the dishonest thrive in our "system".

Suggest altering “it can fail” to “it will fail”.

My plan to destroy the Tesla fleet:

I am for making this a regular test.

You just KNOW Wiley would drive a Tesla (Acme-branded white-label, of course)

(activates FULL SELF DRIVE button) (smiles) (car IMMEDIATELY speeds into a rock wall, crumples like accordian)

That would actually work

There is a parking lot in Portland that has one of these at the Not-an-Exit end…🤔

imagining a pile of ruined cybertrucks at it

bsky.app/profile/nied...

Probabilistic estimation, the most reassuring two words in driving automation technology. It thinks truck nuts are traffic cones, you can extrapolate from there (this is just the funniest example, not necessarily the best)

Trompe-l'œileez nuts

This time it could actually work 😂

Headline does not appear to match accompanying photo OF A TESLA ACTUALLY ON TRAIN TRACKS. "appears to veer"? Weasel words. Fuck off.

I think it's underappreciated that for humans the process is Visual data -> mental model of surroundings -> decision Whereas for a Tesla it's Visual data -> some simple rules -> decision so sometimes the simple rules condemn you or somebody nearby to death.

The Tesla also uses a "model" rather than hard-coded "rules," but it's a statistical model that fundamentally lacks any of the social and psychological insights humans constantly use without even trying while driving. Driving is a highly social task, and statistics only get you so far in those.

This is at the core of AI failures, it uses stats to “infer” the actual answer. Doesn’t use known answers.

TFW you train your model on the boss’ favourite 80s movies backtothefuture.fandom.com/wiki/Railroa...

Railroad crossingbacktothefuture.fandom.com "The car was coasting by the Hilldale housing development now, the place where the railroad tracks crossed the main highway. He [Marty] could hear the crossing gates clanging up ahead as the barriers ...

I should have specified: a set of rules that the programmers don't really know either :)

Also curious if they classify cars by their behavior. Like you'll see an aggressive driver in the rear view and think "that person is going to aggressively try to get in front rather than do a clean merge behind me" and be more prepared when it happens.

One of the fun things about driving in SF is watching the (real) self-driving cars slowly get better at this. They got better about showing intent, about reading other drivers intent, etc. I'd love to know what things they have discovered about real driver behavior and how local it is.

With 40 years of computer systems' and engineering experience, I'd never let a computer take the wheel from me. I will NEVER turn on one of these systems. Yikes.

has anyone asked them to square the stuff about "a self-driving car will drive much more safely than a human" with "a self-driving car should operate with just cameras because vision is how people drive"

Cult members have a ready-to-go reference library of goal-post shifting to match anything FSD/AP fails at. Conveniently ignoring that people's eyes are in fact connected to an organic brain makes their vision-only case much easier.

people use more senses than just vision anyway!

Best line: "it’s not clear whether the vehicle was damaged after driving on the railway line" One might even argue it was 'damaged' before the turn.

That's a pretty spectacular fail right there.

It's a fun low-angle crossing, so I can kind of see what happened, but it's insane that they don't have better checks for how the cars should behave around train tracks that haven't moved in a century bsky.app/profile/mula...

Supposedly Tesla's Autopilot can't handle low-angle railroad crossings. A driver went down the tracks here claiming the car did it on its own abc7news.com/post/tesla-a...

I'm surprised GPS isn't synced to this shit Like without their say so if I invented my own auto pilot pre existing road maps and a compass seem like the first place I'd look....so I can avoid going where roads don't exist

My grandparents survived being hit by a train in their Lincoln Continental back in 1978, they were very lucky. I couldn’t imagine a Tesla surviving a hit like that. I still can not understand Tesla drivers who completely trust a machine with their lives because they are too lazy to actually drive.

Or, because they're so far into the cult of Musk's "genius" that they can't conceive of anything he's involved with being flawed.

Apparently, trains are marvelously new and unknowable to Tesla even though they live on miles of clearly marked paths which frequently cross roads and all are definitely identified on the gps maps Tesla buys to instruct its cunning FSD hive mind. It’s a puzzle.

To believe that Autopilot or any such system can work, you have to believe that driving amongst humans on our current infrastructure can be automated, and it should be obvious that it cannot.

Conspiracy theory: that's why Elon wanted to stop plan for a high speed rail in California

Call in the rail nerds to yell at us that the train tracks are roads too

“Still the best car I’ve ever owned.”

If I were a self-driving system I would rely on GPS and map data to figure out what’s a road, and disable myself when that information is unavailable or unreliable

I qualify this by saying I'm an outlier who has never driven a car, but the whole autopilot thing is mind-boggling. You spent all that money on a vanity car you're too lazy to drive? You GET IN to the VANITY HOOPTY you spent BIG $ ON and can't bring yourself to drive it?

Post