Motorists ‘are being misled by autonomous driving aids’ – report

The Cadillac system requires your eyes on the road. It's harder to game than the systems which attempt to detect your hands on the steering wheel.

Doug DeMuro [1] tested that car a few hours ago and the "Super Cruise" according to him has issues with missing lines. This is a disaster waiting to happen.

[1] https://www.youtube.com/watch?v=NJ6jHgZIEPs

Cruise Automation, developing GMs robotaxis subjects it's human drivers to a month of training before they go on public roads. They're subject to regular safety audits. They keep 2 people in the car at all times. Human monitors take regular breaks. Because it's understood that monitoring a not-yet fully autonomous vehicle is excruciatingly boring, and it's very easy to get distracted or zone out.

The vigilance decrement is a well known and well understood condition of human nature, and Tesla and some of these other companies putting these automated, but not yet autonomous driving features in the public's hands are being willfully ignorant.

Musk may have well come out saying Walter Huang's autopilot death was his own dumb fault because he didn't have adamantium laced bones.

LIDAR has been the best solution to self-driving car vision for a little while now. Interesting to see Cadillac's solution - instead of paying for a LIDAR unit on each car, they pay for a few mapping devices and provide stale data to the car.

> Are there systems in place to automatically invalidate sections of freeway when the government modifies them? Does the government even know that Cadillac is trying to maintain its own map of the freeways?

Indeed, a highly non-optimal situation.

This is why I cringe at self-driving car efforts which depend on extremely high resolution maps, and the companies working on building those maps. Any such map is instantly out of date, so the car needs to have a good enough understanding of its surroundings to be able to ignore the map. Which means either the car is dangerously inadequate, or it doesn't really need the map in the first place. Today's GPS maps are good enough, since a map can only be trusted as a navigation guide anyway.

I have a Model 3 and use Autopilot on a daily basis. I think the key to using it properly is understanding its limitations. I always keep an eye on the road and traffic conditions and anticipate if Autopilot can handle a situation or not so I can be ready take over. I would say it can handle 70-80% of driving without issue, it's the last 20-30% I keep an eye out for when I am using it.

It's made me a better driver, I'm no longer a speed demon or care if people cut me off or get in front of me in traffic. It's like being chauffeured, I am less emotionally involved with the driving process.

I use Subaru's system regularly. You are always in complete control, and it is easy to take over. I liken their lane keep feature to "a strong wind blowing on the side of the vehicle." You just apply a little bit more force than normal and the system immediately shuts off.

I think the systems that are only lane keep/distance based cruise control aren't too bad, because you never feel like it is a fully autonomous system. It is the "uncanny valley" systems, particularly those with misleading marketing, that are getting people into trouble.

I have a Volvo S90 and use both the adaptive cruise control and the steering assist ("Pilot assist" in their language).

Adaptive cruise control I mostly trust. It keeps a set distance behind a car, and that distance varies with speed (think of it as safe braking distance, faster requires more distance).

The only gotcha with adaptive cruise control is if you navigate junctions with it on, i.e. a mini-roundabout on a UK road may be a "straight on" thing and you might see that no-one is going to cross your path, but if you're fixed to a vehicle in front and it happens to turn, then your car will have slowed as it slowed, and then once it detects the obstacle is not there your car will accelerate back to the top speed set in the cruise control. In the UK this could mean going from 20mph to 60mph on a rural road which is a hell of a change.

Steering assist I barely trust. It watches the white lines and seems to favour one side of the vehicle. For clearly and consistently marked lanes of adequate width on only gently curving roads, where you are travelling between 20mph and 70mph, everything is fine.

The gotcha with steering assist is the lack of quality in road surfaces, visibility of lines, missing lane markers, narrow lanes. In these situations the car will have centred me off-centre and I risk being in the oncoming lane on rural roads, or a bend is too dramatic and the car sounds an alarm to get control.

In both cases I do not consider the car as driving itself, I accept these are convenience controls that are limited and so I remain very engaged. Adaptive cruise control is 90% trust worthy but I am wary at lower speeds and small junctions. Steering assist I use only on motorways and dual carriageways now, and never in the slow lane (where motorway ramps, merges, and splits may affect line markings).

One would be wise to not consider the vehicle as being smart.

Scariest moment I've had was actually a different system... collision avoidance.

Edit: There's also an always-on "lane keeping" system which is like the steering assist except it feels like resistance on the steering wheel if you drift over lane markings without having indicated that you intended to. This system I do like.

I have a Honda with the latest Honda Sensing suite.

The lane keep is something that you can turn on over 45mph and it doesn't require you to deactivate it or move the wheel strongly to override it. In practice it feels more like a gentle force keeping you centered in your lane, like driving on rails. I like this feature on highways and it reduces fatigue on long drives. On worse quality roads it can have a bit of a ping-pong motion but these are probably places you shouldn't be using such a system anyway.

I also really like the adaptive cruise which works all the way down to a stop, but I don't rely on it to stop in situations where traffic is coming to a stop quickly. I know that the car would almost definitely stop correctly and with enough space, but the problem is that it doesn't telegraph these intentions early enough for the driver to process and make the call to let the car handle it.

By the time I would realize that my car was approaching the one in front too quickly to stop itself I wouldn't have time to react either, so in these cases I usually override early.

I have never had any indication that the car wouldn't slow down or stop in time and have let it do so on plenty of occaisions, but it isn't a comforting feeling since it doesn't always leave time for the human to process and handle the situation if things were to go wrong. Really this is the same problem as being in a car driven by a human that you don't trust or that likes to brake late.

I've had an all day test of a car with adaptive cruise (Volvo). I didn't like it - it tended to hunt like someone a bit vague was driving. It worked well if others had their speed nailed as if on cruise control too. Concluded I'd use it about as often as regular mechanical cruise (almost never).

I would prefer driving aids to have UI devoted to ensuring one has to keep attention and can't do extremes like the chap photographed alone in the passenger seat of his Tesla at 70. Oh, and marketed accordingly. Until we have full autonomy and the human can be asleep in the back.

I drive a Volvo S90. I only really use the "Pilot Assist" on the highway, and once you've driven it for a while you can get a sense ahead of time what situations it will fail in.

You're right though the biggest issue is still anticipating what other drivers are going to do. You need to override the system to stay out of people's blind spots or else you'll get cut off or worse. When you stay at a steady speed other drivers seem to forget your presence quicker.

I have a model X and I would say while I am driving on the highway in the middle lane I have felt 100% comfortable with auto-pilot and never had to take over. I have driven from Boston to Hilton Head where the car did 90% of the driving and it was amazingly relaxing, literally the reason I bought the car...

There have been a couple situations where I had to take over quickly, but they were in places where I know it might not work out so I was extra attentive. One was a on a interchange between two highways and was a really really tight turn, so I am seeing how well it will do, but also keeping my hand on the stearing wheel ready to take over any second (only tried it once). Another is when passing a city on a highway (NYC, DC, etc) I find that there is always more traffic / crazy drivers / construction / tighter turns so I stay alert and hold the steering wheel ready to take over till I pass the city.

I always set the 7 car buffer in the adaptive cruise control and have no trouble stopping even if there is stopped traffic ahead on the highway.

I have a 2015 Tesla S 70D, so Autopilot 1. I use Autosteer every time I use the car but, of course, not all the time. The way I feel is that together the car and I drive better than either of us can alone. Most of the time the car will alert me to take over long before a problem occurs. Also I will deliberately take control when I see that the car will have difficulty because, for instance, the lane markings have disappeared or there is an oncoming vehicle that is straying close to the centre line. The alerts are unmistakable and both audible and visual and override the entertainment system. Of course on a winding country road one must be more attentive than on a straight, empty, motorway, but that is the case even without Autosteer. I have set the car to maintain the largest separation between it and the car in front which corresponds to about three seconds and to give the earliest warning of collision which means that it quite frequently alerts me and starts braking.

> if u were not in full control from before in a few seconds. It would take that long to even fully understand what is really happening.

But I am paying attention so taking control is really no different and less difficult than what happens to you in a normal car when you have been driving down an empty straight road just lightly holding the steering wheel and a deer or child runs out in front of you.

As a happy EyeSight user, my feeling is that if it takes you any time to understand what is happening you are doing it wrong. These assistive features are great for easing the load of constant mechanical adjustments and mental focus on small details, but do not relieve you of being the driver.

I like to think about it like a ship. Having crew dedicated to rudder, throttle, and navigation doesn't mean the captain gets to take a nap. What it does is remove the distraction of focused attention required by those tasks, so he can be more aware of what is going on around the ship.

Likewise, with a car. With GPS prompting me advance notice of turns, I don't have to try to read every sign going by. With adaptive cruise control tracking the car ahead, I don't have to keep my eyes glued on that car just in case it slows down. And with lanekeeping, I don't need to concentrate so much on lane position. But I'm still the captain. I can't take a nap. I still need to be aware of those things, just not hyper-focused on them. Which frees me up to pay more attention to the cars and other potential threats around me. That greater situational awareness increases safety by improving my ability to anticipate and avoid bad situations, and means I don't need to figure out what's happening if I need to take over.

Perhaps we need a different license system for the niche automaker whose cars are too difficult to operate.

Or... perhaps the niche automaker needs to focus on user interaction design and safe operation.

Why would you want to destroy a real car in this simulation? We know the system is designed to do this, we know it has done it in the real world; it's a waste to use the system to destroy a real car. Plus if you build it right, the fake car exploding into bits is probably pretty dramatic ;)

HN discussion: https://news.ycombinator.com/item?id=17308707

It’s amazing how bullshit stats delivered by Tesla are gospel, but when Monsanto or an oil company cook up self serving nonsense, we’re collectively outraged.

There’s no litigated suit yet, but there are plenty of settlements, including the class action referenced. I’m sure Tesla has a noxious arbitration clause that makes litigation difficult.

Personally, money is a truth maker in these scenarios, and the folks stuck holding the financial bag for Tesla are insurance companies, who are raising Tesla rates for a variety of reasons that are creating losses for them. As this feature is in broader use and Tesla continues to succeed in making their computer’s failures a driver responsibility, those rates will go up more.

Substitute 'The Devil' for 'Capitalism' in these kinds of posts and you have a sermon for sunday school.

I'm going on a tangent (and I'm not trying to troll, I am just as "guilty" as you myself. Worse, in all likelihood)...

I really think that I'm seeing a lot more of "capitalism wants to" or, "globalists are going to" or the good old "the government believes." Abstract concepts personified, deified and vilainised in the way we talk.

Im not against putting things into wider contexts but I do think that this sort of language is leading us astray, more this last year or two then the few before it.

Capitalism, corporatism, neoliberalism and market forces are at play here, so is regulation, centralised decision making.... But, these are abstractions and I don't think they tell us more then the specific does. Let's not just blame whatever abstractions we are used to blaming.

How about (1) people bullshit, especially when they're trying to sell you stuff (2) there may be an uncanny valley type of problem somewhere between cruise control and full autonomy, a level of automation that's not good. It's no longer a driving feature and it's not yet an automated driver (3) laws, driving conventions and common sense will take some time to adapt. (4) we need to make sure we don't let the well understood "people bullshit, especially when they're trying to sell you stuff", especially when their competitors are bullshitting problem lead us into car accidents.