Autonomous cars? I think not

[antigee]

The Guardian reports that Telsa is in court for selling self driving cars with "beta" software.

With the expected (fatal) outcome.

Tesla’s defence, completely without irony, is based on blaming the (supposedly redundant) driver.

"The electric carmaker also claims it was not clear whether autopilot was engaged at the time of the crash."

pretty sure this suggests Tesla only record minimal as required by law data in the EDR (Event Date Recorder) like speed / seat belt compliance - makes no sense for "automated" vehicles

reminded me of this quite close to where I live in Melbourne - no idea on outcome...

https://www.drive.com.au/news/tesla-hit ... autopilot/


"Ms Agrawal had previously told Melbourne Magistrates Court her Tesla Model 3 was being driven in ‘autopilot’ mode at the time of the crash.

In court yesterday, The Age reported, veteran tram driver Glenville Pereira said he was stopped and waiting for a passenger to board when a Tesla vehicle went “whooshing” past and struck the pedestrian."

[Mububban]

Oops

The roll-out of so-called ‘robo-taxis’ in the US has stumbled at another hurdle, after one of the autonomous cars became trapped in an intersection, causing a traffic jam...

...a group of autonomous cars have caused a traffic jam on the streets of a US city, with almost two dozen creating gridlock in Austin, Texas over the weekend.

https://www.drive.com.au/news/autonomou ... -in-texas/

[zebee]

"The electric carmaker also claims it was not clear whether autopilot was engaged at the time of the crash."

https://futurism.com/tesla-nhtsa-autopilot-report

"In the report, the NHTSA spotlights 16 separate crashes, each involving a Tesla vehicle plowing into stopped first responders and highway maintenance vehicles. In the crashes, it claims, records show that the self-driving feature had "aborted vehicle control less than one second prior to the first impact" — a finding that calls supposedly-exonerating crash reports, which Musk himself has a penchant for circulating, into question."

[MichaelB]

Oops

The roll-out of so-called ‘robo-taxis’ in the US has stumbled at another hurdle, after one of the autonomous cars became trapped in an intersection, causing a traffic jam...

...a group of autonomous cars have caused a traffic jam on the streets of a US city, with almost two dozen creating gridlock in Austin, Texas over the weekend.

https://www.drive.com.au/news/autonomou ... -in-texas/

Brilliant !!

[g-boaf]

Maybe the Tesla autonomous driving was enabled on this Model 3?

https://youtu.be/rGLBhrSyVOo?si=BOfHVQ4Olzwo2mVJ

edit: Damn!!

[redsonic]

Article on ABC News website discussing how the autonomous taxis in San Fransisco have eroded people's trust by being unable to cope with emergency vehicles (among other things).

[redsonic]

And this article has a table which shows how Teslas have a tendency to crash into emergency vehicles:

https://www.skynettoday.com/briefs/tesla-investigations

[g-boaf]

Maybe it needs to be left to the conservative manufacturers who take a slow and careful approach to autonomous driving (but still expect the driver to be in the car) who will eventually roll it out properly than leaving it to those who rush to get to market.

[Thoglette]

… take a slow and careful approach to autonomous driving

They’ve been trying since 1939 and, despite the grand promises, there is still nothing to suggest that the fundamental problem (getting computers to behave sensibly in novel situations) is any closer to being solved.

It’s like aircraft: Airbus added some taxiing smarts and was able to get a plane to go from gate to gate autonomously but one needs only to recall the QF incident over Singapore a few years ago as the reason the pilots are not going in my anywhere soon

[g-boaf]

… take a slow and careful approach to autonomous driving

They’ve been trying since 1939 and, despite the grand promises, there is still nothing to suggest that the fundamental problem (getting computers to behave sensibly in novel situations) is any closer to being solved.

It’s like aircraft: Airbus added some taxiing smarts and was able to get a plane to go from gate to gate autonomously but one needs only to recall the QF incident over Singapore a few years ago as the reason the pilots are not going in my anywhere soon

Some manufacturers are doing better than others. If everyone adopts a head in sand attitude then it won’t go anywhere.

I trust them to get it right much more than I trust a “geroff the road” driver to suddenly behave courteously on the road.

And I understand the Qf32 incident, I was at the RAeS lecture with the training Captain who was on the flight.

[zebee]

An article I read recently made a point I hadn't thought of... training data sets need to be country specific.

And in Oz, state specific.

meaning that if you want autonomous cars here they need to be trained here and will that really be cost effective? Will any producer want to manage 7 different datasets to sell maybe 10,000 very expensive cars?

[g-boaf]

An article I read recently made a point I hadn't thought of... training data sets need to be country specific.

And in Oz, state specific.

meaning that if you want autonomous cars here they need to be trained here and will that really be cost effective? Will any producer want to manage 7 different datasets to sell maybe 10,000 very expensive cars?

You didn’t see the autonomous drive S-Class that was running around here with a team of engineers a while ago I suppose?

I happened to spot it by chance, knew exactly what it was. They were gathering data and fine tuning it for our road conditions (and road signs) and eventually that data flows into the shared systems in the other cars sold by the brand.

The levels of autonomous driving eventually gets greater, step by step up until the legislative restrictions.

[Thoglette]

I trust them to get it right much more than I trust a “geroff the road” driver to suddenly behave courteously on the road.

Except those two items aren't related: "geroff" will still attempt to assault you (or have you banned) whether or not cars achieve the ability to sensibly manage situations they aren't trained for.

Autonomy is fine for "know" situations. Lifts, rapid transit, normal flight, maybe even normal freeway driving (as long as "geroff's" mate DGAF is not weaving drunkly between the lanes)

And you can trust engineers as much as you like but as long as the underlying science is not there we're still screwed.

(On .au specific cars and an example of an unexpected situation, look back a few dozen pages for the Kangaroo incident)

[Thoglette]

IEEE Spectrum article Oct 2023 Vol 60 Is 10 What Self-Driving Cars Tell Us About AI Risks (M.L. Cummings)
The article has the following five conclusions.
1. Humans errors in operations are replaced by human errors in programming
2. AI failure modes are hard to predict
3. Probabilistic estimates do not approximate judgement under uncertainty
4. Maintaining AI is just as important as creating AI
5. AI has system level implications that can’t be ignored

Translation
1. To err is human, to really foul up requires a computer
2. It does weird things, randomly.
3. It’s all pattern matching and zero intelligence
4. What it says. (More $$$ each year. Or more weird errors)
5. Does not play well with others

[Mububban]

I genuinely hope autonomous cars can become a safe reality, far too many drivers are terrible at concentration and behaving considerately (we all see it every day on the roads), however I acknowledge that right now, the industry is still in its infancy. Deployment too quickly will simply replace human error with machine error (or human-coded error as described above).





If and when however the systems can be made to work well, and talk with each other, just imagine - the light goes green and ALL the vehicles move immediately and smoothly. Merging would be seamless without egos. Speeding and tailgating would disappear, as would driving under the influence or distracted.

One can dream....

[antigee]

not sure but think this is same program ? ^^^^

https://www.reuters.com/business/autos- ... 023-11-07/

....."We believe autonomous vehicles will transform the way people move around the world, and the Origin is an important part of the AV journey," the spokesperson added.

Cruise, the driverless car unit of General Motors, said last month it would halt operations nationwide after California regulators suspended the robotaxi operator's license, saying self-driving vehicles were a risk to the public.

personally as a short term measure I'd be happy for vehicles to have GPS enabled speed limiters in residential areas with lower speed limits than typical now (50km/hr here in Vic) cheap tech and available now?

meanwhile at 30km range this would get me to the MGC but is probably lacking for the Aus' rural areas where for example getting to a clinic isn't easy:

https://www.cnbc.com/2023/10/13/china-l ... taxis.html

BEIJING — Self-driving air taxis are one step closer to reality in China.

Guangzhou-based Ehang
on Friday said it received an airworthiness “type certificate” from the Civil Aviation Administration of China for its fully autonomous drone, the EH216-S AAV, that carries two human passengers


....Hu said Ehang started doing autonomous aerial flight testing in 2017. There were some vehicle incidents during the early experimentation period, he said, but no big accidents have occurred during subsequent tens of thousands of flights, including overseas.
“Whenever carrying humans, until now, we have maintained a very good safety track record,” he said.

[Thoglette]

If and when however the systems can be made to work well, ….

One can dream....

From ABC Science (James Purtil)
'What you think of AI?' People keep asking this question. Here's five things the experts told me

Sixty years ago, in the glory days of early AI research, some leading experts were convinced that truly intelligent, thinking machines were a decade or two away.

FWIW
I did a couple of “AI” units in the mid eighties and came back to “business intelligence” (AI was a dirty word back then) for a few years in the noughties. Nothing fundamental appears to have changed since. Just the machines (and amounts of money) are much bigger.

[Comedian]

BMW to launch an Autonomous driving system in very limited a places. Sounds like the Mercedes system. "Achieving success by managing expetations".

If we just keep the speed to 60k and less, and don't allow it to work on roads that have pedestrians or animals.. and when the weather is good, and not in tunnels...etc etc.

https://www.drive.com.au/news/bmw-to-la ... many-2024/

[find_bruce]

I can't see if anyone has posted about the 2019 Tesla crash in Florida, where it is alleged the Tesla autopilot failed to detect a turning semi trailer.

Palm Beach county circuit court judge finds ‘reasonable evidence’ Tesla knew self-driving tech was defective

The judge reportedly found that the crash was earily similar to a 2016 crash, also in Florida. As best I can tell the two crashes involved different systems, the 2016 Mobileye & 2019 the Tesla in house system.

The matters considered by the judge included public comments by Musk that " "had a significant effect on the belief about the capabilities of the products." As best I can figure, the finding of "reasonable evidence" is for the matter to go to trial before a jury.

[Comedian]

I can't see if anyone has posted about the 2019 Tesla crash in Florida, where it is alleged the Tesla autopilot failed to detect a turning semi trailer.

Palm Beach county circuit court judge finds ‘reasonable evidence’ Tesla knew self-driving tech was defective

The judge reportedly found that the crash was earily similar to a 2016 crash, also in Florida. As best I can tell the two crashes involved different systems, the 2016 Mobileye & 2019 the Tesla in house system.

The matters considered by the judge included public comments by Musk that " "had a significant effect on the belief about the capabilities of the products." As best I can figure, the finding of "reasonable evidence" is for the matter to go to trial before a jury.

I'm not familiar with the case - but I guess I'm coming to the opinion that it all depends on the metrics.

If autonomous driving systems can never make a mistake ever then I reckon we should probably just give up now.

If success is them making less mistakes than humans for the the same exposure than I guess it's worth continuing. And probably the only way to go forward is for it to be tested in the real world.

The alternative is to do what the Germans are doing.. which is achieving success by tightly managing expectations. IE only allowing it to be used on very limited production models, and in low speed tightly controlled conditions, on certified roads. IE by removing as many potential hazards as they can.

Will be interesting to see how it plays out.

[find_bruce]

As best I can tell it's not actually a criticism of autonomous cars as such but rather the disconnect between the rainbows & sunshine of the marketing & the reality

The judge also cited a 2016 video showing a Tesla vehicle driving without human intervention as a way to market Autopilot. The beginning of the video shows a disclaimer which says the person in the driver’s seat is only there for legal reasons. “The car is driving itself,” it said.

Judge Scott said that “absent from this video is any indication that the video is aspirational or that this technology doesn’t currently exist in the market”, he wrote.

As you say the alternative marketing approach of tightly managing expectations is more realistic & defensible - eg it's not perfect, but better than the average human driver

[g-boaf]

I can't see if anyone has posted about the 2019 Tesla crash in Florida, where it is alleged the Tesla autopilot failed to detect a turning semi trailer.

Palm Beach county circuit court judge finds ‘reasonable evidence’ Tesla knew self-driving tech was defective

The judge reportedly found that the crash was earily similar to a 2016 crash, also in Florida. As best I can tell the two crashes involved different systems, the 2016 Mobileye & 2019 the Tesla in house system.

The matters considered by the judge included public comments by Musk that " "had a significant effect on the belief about the capabilities of the products." As best I can figure, the finding of "reasonable evidence" is for the matter to go to trial before a jury.

I'm not familiar with the case - but I guess I'm coming to the opinion that it all depends on the metrics.

If autonomous driving systems can never make a mistake ever then I reckon we should probably just give up now.

If success is them making less mistakes than humans for the the same exposure than I guess it's worth continuing. And probably the only way to go forward is for it to be tested in the real world.

The alternative is to do what the Germans are doing.. which is achieving success by tightly managing expectations. IE only allowing it to be used on very limited production models, and in low speed tightly controlled conditions, on certified roads. IE by removing as many potential hazards as they can.

Will be interesting to see how it plays out.

Some of the germans are allowing autonomous driving at highway speeds. Depending on what level of autonomous driving, they allow up to 200km/h but the driver must still keep hands on steering wheel otherwise the car will start complaining. That’s a reasonable expectation for a semi autonomous system.

They also use a broader array of sensors which Musk and his lot reckon is nonsense. I disagree, more sensors the better for redundancy in unexpected or difficult conditions. They can all be referenced against each other for a more accurate result.

Better to be conservative and start off gradually.

Volvo is even more conservative!

[Comedian]

Just out of interest.. if the driver has to keep their hands on the wheel and full attention that's Level 2. By dictating the user must remain hands on is just as much managing expectations as governing the speed to 60. It means the driver is still fully responsible and the carmaker doesn't take responsibility.

https://www.faistgroup.com/news/autonom ... es-levels/



EDIT : that level of autonomy is pretty much common place in most cars now.

[g-boaf]

https://www.zf.com/mobile/en/technologi ... iving.html

Mercedes and Audi both have Level 3 in some cars - that's still a pretty big leap considering all the challenges to making it happen from testing and lawmakers.

[Comedian]

https://www.zf.com/mobile/en/technologi ... iving.html

Mercedes and Audi both have Level 3 in some cars - that's still a pretty big leap considering all the challenges to making it happen from testing and lawmakers.

Yes we know that. By adding a lot of conditions. Hence removing as many unknowns as possible. Not in the rain, in not in tunnels, not on roads where any pedestrians or animals can enter, not over 60k, and the Mercedes system doesn't even work in the dark. FFS.

Not to mention there are about 13,000 thousand K's of German road certified. In a country with 644,480k roads.