Tesla "autopilot" disengages shortly before impact 9

[MartinLe]

https://fortune.com/2022/06/10/elon...on-traffic-safety-autonomous-fsd-fatal-probe/

"On Thursday, NHTSA said it had discovered in 16 separate instances when this occurred that Autopilot “aborted vehicle control less than one second prior to the first impact,” suggesting the driver was not prepared to assume full control over the vehicle."

Where I a legislating body, I would very much demand that safety critical software needs to be as transparent as any other safety feature. Which would limit "AI"/machine learning applications in these roles.

 

[SnTMan]

Yeah, well the "Autopilot" and the like has expanded the possible "failure modes" for the human drivers

The problem with sloppy work is that the supply FAR EXCEEDS the demand

 

[dik]

The victim should anticipate a reasonable expectation of something working... else, it shouldn't be available until it is working correctly. It's not an unreasonable expectation... and they are using this technology to sell things... as a main marketing feature.

So strange to see the singularity approaching while the entire planet is rapidly turning into a hellscape. -John Coates

-Dik

 

[TugboatEng]

It's like a car getting hit by a train. Blame is always placed on the crossing, never the victim.

 

[IRstuff]

It's a weird world we live in where the victim must never also be at fault.

Are you just rabble rousing, or do you have hard data? In the case of the Tesla, it's got a barely Level 2 automation, but it's never really sold as such, and I certainly blame the victims for being so easily led astray by marketing of a piece of junk. But, just because the victims share the blame, that doesn't make the system fault free. The bottom line is that Tesla has known about these problems for many years and have continued to fail to overhaul the algorithms; that can't be blame on the victims. If the algorithms did even marginally better, there would be a lot fewer victims for which we can play the blame game. In this last instance, the distance from where the car was still on the highway to the point of impact is about 700 ft, which at 75 mph equates to 6.4 seconds. Given the female driver at age 66, who might have been either inattentive or asleep, it would have taken the entire distance to even come to a complete stop if they were inattentive for more than 2.5 seconds, and any confusion about what was happening and what to do would have only compounded the issues.

There's not much we can do about the victims, since decades of driver education seemingly fails to keep people from speeding or driving recklessly, but we, or more properly, Tesla, can do something about crappy algorithms and poor collision avoidance logic.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[TugboatEng]

Not rabble rousing, I don't think people should be in control of cars in most cases. I really want "autopilot" to work. I just want people to reflect on the fact that this crash was entirely avoidable had driver done anything at all. I'll be curious to see the toxicology report. Once DUI is a factor then the driver is always at fault.

 

[IRstuff]

Tesla's AutoPilot is more than two of SAE's levels away from your desired operation point, Level 4; they've not even nailed down the simplest part of reliable obstacle avoidance, so they're realistically barely at Level 1.

this crash was entirely avoidable had driver done anything at all.

The problem is that even if the driver was fully engaged and alert, that level of alertness would have dropped significantly in less than 30 minutes; the less active driving you have to do, the harder to maintain your alertness.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Alistair_Heaton]

IRstuff what level are aircraft Autopilots/ auto throttles and auto brakes?

All excellent I might add but can still bite your bum if you don't use them correctly or feed them the correct information.

 

[human909]

I just want people to reflect on the fact that this crash was entirely avoidable had driver done anything at all.

I think people here are well aware of that. But it also would not have occurred had the 'autopilot' not failed spectacularly in numerous ways.

Many car manufacturers have sensors and AI that are of similar level to TESLA's, but they don't choose to implement an self driving system like Tesla has due to its risk. The real automation companies that have level 4 vehicles on the roads have FAR better AI and sensors and still choose to limit speeds and locations.

IRstuff what level are aircraft Autopilots/ auto throttles and auto brakes?

SAE levels apply to automobiles, hence the A in SAE.

All excellent I might add but can still bite your bum if you don't use them correctly or feed them the correct information.

Reaction times required are entirely different not to mention the training levels. Comparison to aircraft I don't think is particularly helpful.

 

[Alistair_Heaton]

Well I would disagree on reaction times...

But yes our training levels and knowledge about the systems is far higher.

I wasn't actually meaning to compare them just wondered. As ours throw up funny's occasionally which require manual input to sort out relatively quickly.

They have been talking about pilotless aircraft for years. In fact I was told in a lecture aged 17 33 years ago by the aerospace prof that piloting was a dead end profession. Its relatively easy compared to deal with roads.

 

[human909]

Well I would disagree on reaction times...

I would have thought that in normal steady state flight you don't regularly have reaction time requirements below 3s and reaction time requirements below 1s are not needed. Though I've never flown and if I'm not mistaken you do fly so you can correct me.

Driving regularly requires these sorts of reaction times from people.

 

[Alistair_Heaton]

Its not those phases of flight that are a problem. Although some aircraft it is.

Its 30ft off the ground descending at 12ft per second doing 160 mph you have to be attentive.

This reaction time thing was one of the main issues with MACAS and the 737 MAX. The macas could drive the trim of the aircraft to a point that it became unrecoverable using manual inputs and none power assisted trim inputs. Basically a 737 pilot needs to catch a trim run way in under 3 seconds and not let the speed increase to outside the 10 knot window they can trim manually.

 

[human909]

Its not those phases of flight that are a problem. Although some aircraft it is.

Its 30ft off the ground descending at 12ft per second doing 160 mph you have to be attentive.

Exactly my point. That is a know point in the flight phase. You aren't cruising along halfway across the Atlantic and have to perform a landing within 3 seconds.

Tesla's 'autopilot' expects you to have high reaction and decision making abilities at all times while happily shouldering the work load 99% of the time. 1% of the time that it fails it can fail quite badly and unexpectedly and requires a user to intervene quickly to prevent disaster. An attentive user in this case would have likely prevented this accident, this is almost certainly why there aren't MANY more of these accidents.

This reaction time thing was one of the main issues with MACAS and the 737 MAX. The macas could drive the trim of the aircraft to a point that it became unrecoverable using manual inputs and none power assisted trim inputs. Basically a 737 pilot needs to catch a trim run way in under 3 seconds and not let the speed increase to outside the 10 knot window they can trim manually.

Yes and we all know how that turned out. But that is hardly relevant because that is just as flawed as Tesla's autopilot.

 

[Alistair_Heaton]

Exactly and there is much much less variables that an aircraft automation system has to deal with.

If they haven't cracked it getting on for 90 years worth of development with aircraft automation. How is anyone even considering it for cars with thier vastly increased number of variables and risk factors and shear volume of data that needs processing.

Its the whole concept of what is a humans reaction time to failure of automatics. I suspect Tesla is using Boeings 1960's criteria for fail passive reactions. Most people will struggle to process the situation data to even begin reacting in an appropriate manner.

There is a huge subject called Threat Error management or TEM for short. Which then links into Crew resource management which deals with human performance in aviation. As far as i can see car autopilots fail miserably in both aspects in a public uncontrolled environment.

 

[LittleInch]

I think just about anyone on these pages knows that the Tesla system is significantly flawed and should never have been marketed as "auto pilot" implying vastly greater levels of authority than it should.

In this instance if you track back along the high way, the inner lane with the continuous white line on the right splits into two with the 4th Lane becoming the slip road. So it kind of makes sense that if the car was in the RH lane then it would move over a bit. But then there is quite a sharp turn right when the white line becomes continuous on both sides. You can only guess that once the car was a few metres off the road where the GPS thought it should be that it then tried to get back onto the highway. Unfortunately there was a rather large trailer in the way. We don't know what other trucks were parked there so not possible to see what the system thought was a good idea, but these systems are just not ready to be used in this way.

Maybe there was another truck hiding the second one and then once it made the turn ran out of time before this other truck it couldn't see was stationary. All at 60-70 mph....

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

 

[Lou Scannon]

Actually, SAE embraces all forms of mobility engineering.

"Schiefgehen wird, was schiefgehen kann" - das Murphygesetz

 

[IRstuff]

Actually, SAE embraces all forms of mobility engineering.

Ditto, a go-to document I used to cite regularly was SAE-AIR-1168-4, which covered icing and deicing on aircraft

You can only guess that once the car was a few metres off the road where the GPS thought it should be that it then tried to get back onto the highway. Unfortunately there was a rather large trailer in the way.

Not sure if that's the case, since at least Google Maps "knows" about rest areas. I personally don't know what Tesla uses or how they integrate it with both the lane following and/or collision avoidance, but it's pretty clear if the car was under navigation control, its integration with lane following/collision avoidance is 3rd rate, at best. The navigator should keep the lane following from FUBAR'ing that badly, and the likewise, navigator shouldn't steer the car into a parking spot and collision avoidance should have stopped the car before collision. This is really basic stuff that shouldn't get screwed up so badly.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[LittleInch]

Well I don't know about anyone else, but occasionally my sat nav decides I'm actually on the minor road running right along side the main road and starts going nuts telling me to turn right / do a u turn / "recalculating". GPS is pretty good but it's accuracy can vary up to 5 to 10m on occasion if it looses a satellite or two / runs through some heavily wooded areas.

But I agree, wherever the car thought it needed to go it should have worked out that the big stationary object in front of it needed to be avoided.

My only guess is that the truck it actually hit was hidden by another truck parked further north. Maybe another truck was in the way trying to park so the car swerved to avoid it and hit something else instead?

But it should also have recognised it was in a service area and slowed down as it entered.

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

 

[Alistair_Heaton]

SAE-AIR-1168-4, which covered icing and deicing on aircraft

I knew I had seen SAE somewhere to do with aircraft. Must be the FAA deicing tables.

 

[btrueblood]

"SAE levels apply to automobiles, hence the A in SAE"

Society of Automotive Engineering - i.e. any self-powered mobile device, aircraft and spacecraft included.

As a young aero grad student, we published (or at least attempted to) in this journal as well as the AIAA journal(s):

https://www.sae.org/publications/collections/content/e-journal-01/

 

[IRstuff]

SAE-AIR is an entire series of aero-related documents; below are the ones I have PDFs of

SAE AIR 1064C Brake Dynamics.pdf
SAE AIR 1075A Barometry for Altimeter Calibration.pdf
SAE AIR 1093 Numeral, Letter and Symbol Dimensions for Aircra.pdf
SAE AIR 1102A Transparent Area Washing Systems for Aircraft.pdf
SAE AIR 1106A Some Factors Affecting Visibility of Aircraft Navigation and Anticollision Lights.pdf
SAE AIR 1151 Flight Compartment Glare.pdf
SAE AIR 1168-1 Thermodynamics of Incompressible and Compressible Fluid Flow.pdf
SAE AIR 1168-10 Thermophysical Characteristics of Working Fluids and Heat Transfer Fluids.pdf
SAE AIR 1168-13 Spacecraft Equipment Environmental Control.pdf
SAE AIR 1168-14 Spacecraft Life Support Systems.pdf
SAE AIR 1168-2 Heat and Mass Transfer and Air-Water Mixtures.pdf
SAE AIR 1168-3 Aerothermodynamic Systems Engineering and Design.pdf
SAE AIR 1168-4 Ice, Rain, Fog, and Frost Protection.pdf
SAE AIR 1168-5 Aerothermodynamic Test Instrumentation and Measurement.pdf
SAE AIR 1168-6 Equipment Components, Equipment Cooling System Design, and Temperature Control System Design.pdf
SAE AIR 1168-7 Aerospace Pressurization System Design.pdf
SAE AIR 1168-8 Aircraft Fuel Weight Penalty Due to Air Condi.pdf
SAE AIR 1168-9 Thermophysical Properties of the Natural Environment, Gases, Liquids, and Solids.pdf
SAE AIR 1221 Electromagnetic Compatibility (EMC) System Design Checklist.pdf
SAE AIR 1608 Estimation of Total Error in Altimetry.pdf
SAE AIR 1609 Aircraft Humidification.pdf
SAE AIR 1657A Handbook of Hydraulic Metric Calculations.pdf
SAE AIR 1667 Rotor Blade Electrothermal Ice Protection Design Considerations.pdf
SAE AIR 1678 Uncertainty of In-Flight Thrust Determination.pdf
SAE AIR 1703 In-Flight Thrust Determination.pdf
SAE AIR 1758 Limits and Fits - International Metric Tolerance System.pdf
SAE AIR 1780A Aircraft Flotation Analysis.pdf
SAE AIR 1812A Environmental Control Systems Life Cycle Cost.pdf
SAE AIR 1823A Engineering Analysis System (EASY)( Computer Program for Dynamic Analysis of Aircraft ECS.pdf
SAE AIR 1845 Procedure for the Calculation of Airplane Noise .pdf
SAE AIR 1855A Actuation System Data Summary for Missiles and Launch Vehicles.pdf
SAE AIR 1939 Aircraft Engine Life Cycle Cost Guide.pdf
SAE AIR 2000E Aerospace Fluid System Standards.pdf
SAE AIR 4013A Multiplex Data Bus Networks for Mil-Std-1760A Stores.pdf
SAE AIR 4015A Icing Technology Bibliography.pdf
SAE AIR 4092A PTFE Melt Phenomenon for High Pressure Hoses.pdf
SAE AIR 4093 Compendium of Gas Properties R(1994).pdf
SAE AIR 4094 Aircraft Flight Control Systems.pdf
SAE AIR 4298 Impulse Test Machine.pdf
SAE AIR 4359 Effects of Hanging Loads.pdf
SAE AIR 4367 Aircraft Ice Detectors and Icing Rate Measuring Instruments.pdf
SAE AIR 4548A Real-Time Modeling Methods for Gas Turbine Engi.pdf
SAE AIR 4827 Modeling Techniques for Jet Engine Test Cell Aer.pdf
SAE AIR 4845 FMECA Process.pdf
SAE AIR 4911 Requirements Document for Sensor--Video Intercon.pdf
SAE AIR 5006-2 Volume II Probabilistic Design and Analysis Methods for Solid Rocket Boosters.pdf
SAE AIR 5020 Time-Dependent In-Flight Thrust Determination.pdf
SAE AIR 5022 Reliability and Safety Process Integration.pdf
SAE AIR 5080 Probabilistic Methods into the Design Process.pdf
SAE AIR 5086 Perceptions and Limitations Inhibiting the Appli.pdf
SAE AIR 5109 Applications of Probabilistic Methods.pdf
SAE AIR 5145 Directory of Databases Part I - Whole Body Anthr.pdf
SAE AIR 5396 Characterization of Aircraft Icing Conditions.pdf
SAE AIR 5689 Light Transmitting Glass Covers for Exterior Aircraft Lighting.pdf
SAE AIR 809 Metal Dimensional Change with Temperature.pdf
SAE AIR 818D Aircraft Instrument and Instrument System Standards.pdf


TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm