Tesla Full Self-Driving videos prompt California's DMV to rethink policy on accidents California’s Department of Motor Vehicles said it’s “revisiting” its opinion of whether Tesla’s so-called Full Self-Driving feature needs more oversight after a series of videos demonstrate how the technology can be dangerous. “Recent software updates, videos showing dangerous use of that technology, open investigations by the …
"There was a Top Gear episode where they interviewed the woman who did the design of the signs."
Margaret Calvert. If you have access to BBC iPlayer it is currently available as Series 14, episode 7. She also recently featured in Secrets of the Museum, the programme about the V&A, but that's not currently available.
Or see:
http://britishroadsignproject.co.uk/jock-kinneir-margaret-calvert/
Incidentally, the brown background for tourist attractions (introduced in the 1980s) has more recently been adopted in other European countries.
Not all of them are. STOP being one example. GIVE way and most roadworks signs are other examples.
But I do find that the work done in the 60s which informed signage and communications for the next 50 years has produced the best signs in the world. Unfortunately the gains we made in this are slowly being eroded as the complexity has increased and the "KISS" minimalist principle has been sidelined in favour of nannying and the increasingly litigious society here.
"You seem to be writing a computer program, want some help?"
I will have nightmares. I still cannot get used to Xcode's habit of automatically inserting a closing '}' when I start a clause in 'C', god help me if my computer starts talking to me.
I know, I'm old and was superseded decades ago. I'll just talk amongst myself.
"I still cannot get used to Xcode's habit of automatically inserting a closing '}' when I start a clause in 'C',"
Sounds like a good idea, providing the opening brace is:
a) Always vertically aligned with matching closing brace.
b) Never allowed to reside at the end of a line.
( Sound of hatch being screwed shut )
It is a good idea (hence the upvote from me), it is just that I learnt BASIC when I were a nipper (I got an 'A' in the very first 'O'-level Computer studies option in nineteen-seventy-something), and these new-fangled WYSIWYG and predictive interfaces still amaze me. (Also do not get me started on 'autocomplete' and 'autocorrect'.)
Modifying existing code... yes. Many IDEs do this, and if you are say moving a block of code into a conditional branch, then this feature can be a royal pain in the butt. You know which block of code you want to enclose, and you would rather put the } in yourself, but even if you put one of those FIRST, before the conditional or the {, then it still puts in a { }, meaning you have to remove the extra }.
Because it's useful in the other scenario and a faff to turn on and off through preferences. You start a conditional block and get typing and lo! Your code is nicely bracketed.
Now if the IDE recognised, say /} as end of block which matched with a { typed afterwards or it recognised a selected block of text and didn't replace with { when you typed { but did a {...} instead then we would be laughing. There are IDEs that will enclose a highlighted block in different sorts of inverted commas but not {} or ().
wordsmith: "Why don’t you just go into preferences and switch it off?"
O M G!
You mean I can actually turn it off?
Honestly that had not occurred to me.* I am such an idiot.
I hang my head in shame and humiliation.
A good job I was never, well only briefly, actually employed to write a program.**
*True, it really had not occurred to me, this is not a piss-take, hence the 'face-palm' icon rather than the 'joke alert' icon.
**It was implementing an HMG flow-chart like process to assess IT systems, and the code is awful as I was writing it in C++ for Windows, because someone else in the company wanted C++ on their home computer and 'accidentally' copied the installation disks... (This was a long time ago and sadly he died recently, so please, no writs or summonses, please.)
Don't forget that you will soon have to fork out a monthly subscription for the [cough][cough] service.
Tesla has been increasing its prices rapidly in the past year. They say that it is to quash demand. Bovine Excrement to that.
With their Texas and Belin plants about to open they will have plenty of capacity to fill their backlog.
Tesla's are cheaply built basic cars that you are paying £20000 for the name.
There was quite a substantial amount of R&D that went into the battery and motor technologies alone.
To suggest that you are paying more than double the representative cost of the car is rather far fetched.
Tesla's are cheaply built basic cars that you are paying £20000 for the name.
People think a £50k Tesla is a £50k car and expect something Mercedes, or high-end Audi in quality. It's actually a £20k car wrapped round a £30k battery, which is why overall you get something more Ford or Hyundai. Perfectly competent, of course, but nowhere near the luxury or quality you might expect for the money.
A while back I joined a road from a T junction and noticed two Audis about 1/4 mile away racing each other approaching very rapidly from behind. I was planning to turn right (in UK so driving on left of road) in a few hundred yards and there was another car coming in the other direction ... which would pass me just as a got to the junction. A plan formed. The rest of the world was tuned out. It helped that I was in my old man's Rover 75.
As the Audis tailgated me, desparate for the oncoming car to go past so they could overtake, I bided my time. JUST as the oncoming car went past, as the boy racers revved their little engines, I indicated right.
A peal of brakes and a lot of honking ensued from behind. They had to wait till I turned right.
Win win win.
"and better than Mercedes"
Bjorn Nyland did a review on an EV Mercedes sedan where he raved the whole time. I hope he had a towel under him to preserve the seat material. In the case of the Merc, it was a $100k car with a $40k battery fitted.
If I were to buy right now and could even find a second hand Hyundai Kona EV or Kia Niro EV, that's what I'd get. The very traditional controls are a big plus for me. I hate the spartan interior with an iPad glued to the center of the dash that everybody seems to be copying Tesla on. I'll likely wind up with a Chevy Bolt as that's a perfect fit for my needs right now. It all depends on my budget and what's available when I buy. I envy the choices in Europe and Asia.
No interest in any self-driving fantasies.
Audi quality, yes the current Tesla's are exactly Audi quality. Stylish plastic everywhere and an electrical system that has a 2% cumulative chance to burn the car and your garage down in the middle of the night.
I've driven both extensively, and I own neither because I actually prefer my mid-market korean econobox.
That said, the model S exceeded my expectations. EVERY friend of mine that bought an Audi had buyers remorse(and a sore backside, those lads like their suspension rough...)
I did recently take a ride with a friend who had just bought a new Audi... you know that "new car smell" that's so sought after? A slightly heady mix of leather, carpet, light machine oil and lightly spiced rich fruit (I've no idea what that actually comes from)? Well this one reminded me of the olfactory component of the OOBE you get when taking the lid off a new Airfix kit.
So I saw a Tesla turn onto the end of my parent's street... and keep turning, until it hopped the curb and was about to the sidewalk, at which point the driver finally grabbed the wheel and swerved it back onto the road. And that was not in snow-covered conditions as we have now, that was on a 100% clear road, dry, broad daylight, and 2 or 3 inch curbs with grass on both sides of the road.. i.e. not some low-contrast or poor visibility situation.
My friend has a Tesla, they commented they don't even use the automatic cruise control, one of they highways where they live in Colorado, they found as it approaches where this bridge with fencing passes over the highway that it slams on the brakes! Despite the bridge and fencing being probably 20 feet above the road (I don't know exactly how high above, I haven't seen it) they assume it's misinterpreting the fencing as a solid object on the road. Yikes!
Guessing one of 2 things. Map data not being very accurate, or accurate enough. This used to be a bane of my life with customers who wanted kmz files showing fibre routing. Luckily I had some papers comparing proper survey data accuracy to Google and similar map accuracy. It makes life a whole lot easier for network designers & planners if clients just say who they want a route diverse from.
Or the car's 'AI' was having a bad day, and couldn't recognise road markings, confused by street clutter etc. I've heard it's possible to do this by just sticking a few wheelie bins close to the kerb either side of a junction. Works best when it's a narrow side street. Allegedly.
It's the latter.
Map data is irrelevant, in that it is static data, whereas hazards can be both static and dynamic. The map may very well say something like, "There's a bridge over this road, it's got quite tall side-fences on it". However, that does not mean there isn't a jack-knifed lorry stuck underneath it. So the car has to recognise hazards present / absent at the bridge, independent from the map data.
The "street clutter" thing is going to be interesting. Already here in the UK there's people in the self-driving car industry lobbying for laws to "protect" self driving cars from being harassed by pedestrians / cyclists who behave in a manner designed to exploit the self-driving car's obligation to not hit them.
The problem with such lobbying is that with the state of the technology at the moment, "protect from harassment" basically translates into "no cyclists / pedestrians / motorbikes / emergency vehicles / horses / horse & carts / stray animals / road works / ordinary cars / shiny shop windows / etc.". Good luck to which ever government tries passing and enforcing a law like that!
The "street clutter" thing is going to be interesting.
Until the car systems can apply a "is this plausible" test it will continue to be so. Just scanning the environment for obstacles isn't sufficient, it needs to understand those obstacles in context, and none of the much-touted "AI" systems are even close.
It's like the hacks people play with street signs. If you're driving though a residential street with pedestrians around and you see a speed limit sign showing "80" your first thought will be "that can't be right, someone's been messing about with black tape on a 30 sign". A Tesla just goes "I can do 80MPH! Wheee!"
A so-called "high-tech road" that announces speed limits, schools, kindergarten and other high pedestrian might be useful to more than just elf-driving vehicles (not self, there is a pixie in there, hence the "elf"). Data in real time.
Plod, Ambo and Firey can announce themselves to drivers in the vicinity. Perhaps a red light that actually cuts the engine until it turns amber, on the way to green. Countdown to a red to conserve fuel (dead dino and electrickery) And more (ok, I ran out of examples).
Two problems, at least. 1 - keeping this data up to date. Do you need to know about the school during school holidays for example? Perhaps, perhaps not. Who keeps it up to date - Local council? The state? Private company via annual sub? 2 - validating the signal. Signed and encrypted message packets broadcast from the lampposts? Three problems. Idiots that treat each of these as quests al la GTA VI.
I know we have traffic broadcasts and the various map providers have these details already, but "high-tech" road is about having this integrated and updated in real-time.
Another consideration to embedding metadata into the road is how to maintain this metadata if (a) information changes, such as new speed limit assignments or a school opening or closing, or (b) the things that present the data get destroyed for some reason. It has some potential, but it wouldn't be a set-and-forget kind of thing; it would need upkeep which costs time and money.
I agree on both points, and most of the existing navigation software already has access to the database of speed limits, even without self drive. Camera's could read the signs, but the car (navigation or otherwise) should only be using that to REDUCE speeds, not increase them over the base speed of the road.
It might also help if the damn things remember were they have had trouble in the past, and just tell you take the wheel whenever you are approaching one of them.
I think we are going to need to deal with improved road markings and vehicle metadata to address some of these concerns, and adding an easily machine readable marking placard to vehicles that tend to stop in the middle of the dam road is also common sense. I'd also tag a suitable fine on any merry prankster misusing them with a snapshot from the car's cameras being sufficient to enforce a citation.
Clearly, reliable self drive on city streets is a ways off, so we should probably consider marking up the streets we want to route self drive traffic on so that the cars don't screw up as much, and if they do, pedestrians, bikes, dogwalkers et al can avoid them. Not such a hassle if the street over is car free.
"remember were they have had trouble in the past"
Stop treating them like they've actually got any intelligence at all. What do you mean by 'had trouble'? Those Waymo cars recognise when they're in trouble... it's when they haven't made any progress for 5 minutes when they have an active fare. Meanwhile 4 minutes 30 seconds ago the active fare has been on the blower to Waymo Central complaining that their car has got stuck. Having said all that, I am impressed with what Waymo have managed to achieve.
Also, street markings are fine on brightly sun lit Californian highways more or less devoid of trees, rainfall, leaves, accumulations of litter, snow etc. For 50% of the year road markings in the UK are invisible or barely visible.
You are 100% wrong about Tesla using only GPS with no video. No idea where you got that information, but Tesla has been 100% Video with no LIDAR used ***in combination*** with GPS.
Just do a simple internet search and you’ll find true information about this. One of the posts above posits about pranksters and speed limit signs. Gee, maybe that could be tested?
https://www.extremetech.com/extreme/306346-researchers-tape-speed-limit-sign-to-make-teslas-accelerate-to-85-mph
That's weird, my Opel has sign detection. Hell, some of them even allow you to set cruise control based on the signs. As far as I can tell, it was first developed by GM and landed in the Opel/Vauxhall Insignia before others, so it would be bizarre if Ford had a patent on it.
And that's WITHOUT the legitimate highways departments who simply can't get signage correct.
For example (after I complained about it at least six times they've now fixed this I see on street view when I went to check just now) the "keep left" downwards diagonal white arrow on blue background that directed vehicles to drive onto the pavement at the junction of Tollington Road and Holloway Road.
And at the junction of Roman Road and Watford Road in Elstree where there on the traffic lights there's a no right turn sign, a bus lane sign and no right turning green arrow on the traffic lights for the bus to turn right through its bus gate. This is another I complained about and this is what they came up with AFTER I nagged them about it. You can use the timeline view to see the dog's breakfast that it was before then.
And at the end of Watford High Street where the bus routes go straight ahead, but they must also turn right and are banned by the signage on the traffic lights. Thankfully the bus drivers have wetware which allowed them to ignore the traffic signs and continue driving the bus route they had been doing for some years previous.
And as for the much abused pass either side TO REACH SAME DESTINATION sign that has countless times been used as if it means just pass either side of a rather obvious obstruction in the road...
So basically if the DESIGNERS of road schemes can't even communicate the rules they want to employ for certain situations using a nationwide and standard set of icons, how can they even contemplate an AI to be able to understand the rules? And they still think that self-piloting vehicles are a viable thing of the future??!!
to build a better system where the cars report the rogue signage and there are actual consequences if they haven't been updated after a reasonable period of time. I'd also block the local police from issuing citations until it's fixed.
While I appreciate your concerns, our local government is similarly inept, but they can remain so because there is neither transparency or accountability forcing them to either do their damn job or be fired. If the database of signage faults is public, it will be a bit harder to avoid explaining why the person responsible isn't fixing the problem.
Also, I really doubt the existing self drive will do well in that kind of street environment, and should be dropping out of self drive and into "watch the driver like a hawk and keep them from driving into that obstruction you mentioned" mode instead. Or refusing to drive 85 mph on a pedestrian walkway.
I believe that the UK (or at least England) is planning legislation to protect the most vulnerable road users even more, that means basically pedestrians and cyclists, and car drivers (whether human or AI 'enabled') will have to give them priority.
So when we all drive, sorry, ride in, self driving vehicles, Extinction Rebellion and Insulate the UK won't have to glue themselves to anything, they can just wait for the next Tesla to come along and step out in front of it.
I believe that the UK (or at least England) is planning legislation to protect the most vulnerable road users even more, that means basically pedestrians and cyclists, and car drivers (whether human or AI 'enabled') will have to give them priority.
The new rules are due to come into force on the 29th of January.
The changes are documented in
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/1037306/table-of-change-to-the-highway-code.pdf
will be meaningless when (not if) the Cybertruck starts appearing on our roads. The 'tank' will give occupants a sense of invulnerability.
I am sure that it won't be long before some jerk in one tries to find out just how much of a tank they are. Cyclists and Pedestrians won't stand a chance.
I'm feeling rather good that as a driver the only thing in there that will require a change to how I drive is giving pedestrians priority at junctions. Which I already do if they're in the road, just not necessarily while they're waiting.
But that's been normal in Germany for decades so I'm used to it from there; it'll be trivial to adapt here.
The thing that annoys me is the acceptance that cyclists can just bundle down the left of whatever vehicle they choose. Quite why the Highway Code (and associated laws) don't tell them to never go down the left of a vehicle indicating left I don't know. Maybe the authors want dead cyclists.
"The thing that annoys me is the acceptance that cyclists can just bundle down the left of whatever vehicle they choose. Quite why the Highway Code (and associated laws) don't tell them to never go down the left of a vehicle indicating left I don't know. Maybe the authors want dead cyclists."
No sane cyclist would - unfortunately years of farcility design have tried to force that decision on cyclists.
Take a look at the legislation around the advance cycle boxes at traffic lights. I have one near me where the access into it (via the cycle gutter) is on the far left, so even if I want to turn right I have to access it and then cut across multiple lanes of traffic.. and the traffic lights for straight ahead could of course turn green at any point during that pointless manoeuvre. All they needed to do was make an exception for pedal cycles - don't need to stop at the first line, still required to stop before the second line.
Perhaps I misread but I thought they do - explicitly stating that where there's an advance box available, cyclists may go through the first white line (which cars must stop at) but must stop at the second until the light turns green.
Of course, no road markings require a cyclist to pass a vehicle waiting to turn. The cyclist could pause and wait behind it, just as other road users do.
Nope...
TSRGD2002 Part 1, Section 5, Regulation 43(2):
Where the road marking shown in diagram 1001.2 has been placed in conjunction with light signals, “stop line” in relation to those light signals means—
(a)the first stop line, in the case of a vehicle (other than a pedal cycle proceeding in the cycle lane) which has not proceeded beyond that line; or
(b)the second stop line, in the case of a vehicle which has proceeded beyond the first stop line or of a pedal cycle proceeding in the cycle lane.
I agree that most cyclists quite reasonably ignore this idiotic legislation, but that's what the law actually says.
Oh! Is that still the draft version of the new Highway Code?
You see I did feedback on that when it was up for consultation... and the main issue I found was that they made no distinction between a signalled and an un-signalled junction or junctions with or without signals for pedestrians, or how a driver is supposed to know if a pedestrian has signals or not now that they've taken to tucking the signals away on little boxes only visible to the pedestrian.
The way that they've written it, if I'm in a vehicle, turning left, and I have a green light and there's a pedestrian on the pavement on my left or even on the opposite side of the side road, then I have to stop and give way to them even if they might have a red man on that little box on the pole, and even if it means that I'll then run the risk of the lights changing and me being in conflict with other vehicles who were previously being held on red.
Seems ridiculously loosely written to me. If they're going to rely on the application of common sense then they might as well have cut around 2,500 words out of it and just out in a reminder about being courteous and HGV > car > bike > pedestrian as far as ability to inflict damage and injury goes.
My god, what a nightmare! I'm not paying for an autonomous vehicle if it's going to give way to plebs!
We need some kind of "net worth" scanner so it can brake for high value individuals but ignore those who can't be bothered to improve themselves.
From the LA Times: "Unlike the other companies, Tesla is doing without trained test drivers. Participants in the Full Self-Driving beta have paid $10,000 for the privilege — soon to be raised to $12,000"
So not only are they not reporting hazardous incidents, but they're actually getting members of the public to pay big bucks to expose themselves to them.
My son is a senior manager (VP Research and Operations) of an autonomous transport company. It’s all he’s done since university, started in development, promoted to run research and ops with famous company #1, moved on to famous only in the US company #2 that has no human passengers, then global famous company #3 that has no human passengers.
Has hired people from Tesla & relayed a few (WTF?!?) tidbits to me they confidentially told him.
Bottom line is there is no way possible for video only to be safe. A combination of at minimum LIDAR and video are required. Musk keeps poo pooing LIDAR due to cost, but there are some in the larger autonomous industry willing to testify in court that video only is a ridiculously dangerous approach.
As someone above noted, why would you buy an autonomous vehicle to ride in? I’m against the “assistive“ highway tech off GM for example because it encourages inattentive behavior, but at least can understand the motivation of the traveling salesman as I frequently drove 300-400 miles/day 40 yrs ago while in sales.
Bottom line is there is no way possible for video only to be safe. A combination of at minimum LIDAR and video are required.
Remind me. Of the 32 million cars on the road in the UK, how many are equipped with LIDAR and how many are driven around based on video input only?
Of course they're not particularly safe ...
Better.
Video only relies on a huge amount of contextual information - bridges are made of these kinds of stuff and exist in these kinds of places (cuttings etc), Audi drivers will cut you up, Mercs are probably going to climb into your boot etc.
I'm certain that it's possible to make an autonomous vehicle that only has video (and perhaps audio) external sensors, because there are tens of thousands of examples of excellent ones.
However, it's clear that it's very difficult, because there are also tens of thousands of examples of abysmally dangerous ones - and many millions of mediocre autonomous agents.
They also seem to take about 30-40 years to gain sufficient training data. 17-20 years is borderline.
Let's not play the false choice game with Elon and the shills for the LIDAR companies again. They can be both wrong, and this case they are. Adding LIDAR on top of cameras dosen't make things magically work all of a sudden. And while an LIDARless approach is perfectly valid, it comes with the need to accept that when the cameras can't work out what's happening, you need to not try to pretend you can keep self driving with them.
The problem with a hybrid system that uses LIDAR or another non-visual technology is that you then have to correlate and deconflict the results returned by those systems in real time. If the cameras and LIDAR/radar/sonar don't agree, you can't just assume that the LIDAR is always right. It faults more often and in more dangerous situations than the cameras do. Not one of the companies trying to do self driving has even tried running with a pure LIDAR solution.
They are selling the idea that only LIDAR can save self driving when in fact, a LIDAR equipped car that hasn't also solved the camera based machine vision problems isn't automatically safer then the camera based car. It becomes a matter of the vehicles systems making correct decisions on conflicting input from it's systems. If you hand that decision to a ML system it will surprise you by making wildly unpredictable choices.
The LIDAR equipped cars are still running over pedestrians an accelerating towards bicyclists. They are also more likely to go into sudden breaking when nothing is in the road ahead of them. The kind of muti-sensor fusion they are attempting is still an unsolved problem in computer science. The F-35 has the same problems. The solutions they have amount to mostly working most of the time. Until that changes, requiring LIDAR or any other situational sensor system isn't as important as understanding the limits of what the system can safely do and sticking to them. Maybe that means in both cases that the vehicles only allow self drive on stretches of road that have been worked out to present an acceptable level of risk, and under good conditions.
If that means peoples expensive toys can't drive themselves from door to door so be it. If you don't want to drive that bad, pay someone else to do it.
Just like with multiple cameras with overlapping views you have to have a system that collates the information and creates a model from the various inputs.
What's your suggestion? Using less info will give a clearer picture cos no contradictions and no tricky decisions to make?
I disagree. The problem does exist: it's that meatbags driving means a high casualty rate (particularly in the US for some reason). I'm sure we've all had to dodge fuckwits on the road who frankly shouldn't be allowed to drive.
However, solving this problem by using "autonomous" cars still has a long way to go.
Still, it doesn't have to be perfect - just better than humans.
Meatbags driving has an incredibly LOW casualty rate comparatively. It's just not zero, and that's a laudable target. I reckon if they'd put all the resources that they have done into self-driving into taking aggressive, idiotic and inconsiderate drivers off the road permanently then they'd have had a lot more impact on those casualty figures.
I still stand by the "if the self driving cars are killing less people per mile than people driving them" yardstick, but that doesn't mean we shouldn't raise the bar for human drivers too. Here stateside the drivers test is a joke, and no meaningful training before, or ongoing training after, is required. Even if you cause multiple wrecks, if you don't have a medical condition, or aren't a senior citizen, you don't even have to re-do the driving test.
Even requiring a weekend long crash course in a video game simulator that forced you to react to common road accident situations every 5 years or so would probably save thousands of lives in the US.
I can think of quite a few cases where self-driving would be useful.
People with disabilities need to travel for work and shopping. Lots of disabled people who could work are unemployed because of a disability that limits their ability to drive, and public transit is often not a viable alternative.
Old folks. In the US, driving is often a key part of being independent. As people get too old to drive capably, there's no great transition for them. Self-driving cars could fill that gap.
Fatigue. I've read some studies that e.g. nurses have a higher risk commuting home after a 24-hour shift. A self-driving car could save lives.
There does seem to be disagreement about the necessity of LIDAR. LIDAR generates a point cloud, which is then processed to develop a model of the car's environment. Computer vision can achieve the same outcome and has the benefit of being able to recognise objects only visible to drivers - e.g. speed limit signs, road markings etc. Depth perception requires inference - this is a moving object, this is stationary, this is a parked car etc.
LIDAR does have the advantage of directly measuring distance but It has it's own problems - coverage/field of view, range, compatibility with rain/mist/fog.
You could argue that both are necessary, but the problem then becomes fusing the sensor outputs and dealing with disagreements. Personally, I think Tesla is on the right track but has a long (infinite?) way to go before achieving anything beyond L3 autonomy. So does Mobileye and every other autonomous driving solution in development.
The press insists on calling this Full Self Driving, it's not. It's a BETA that requires human monitoring.
They often imply that all Teslas are self driving, they are not. The BETA is only available to a relatively small number of testers.
They often claim countless FSD crashes. Well they are countless, because to date there are zero such crashes.
The Phantom Braking issue with enhanced cruise control (which Tesla confusingly calls "Auto Pilot") is indeed a significant issue on two lane roads.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
