Self driving cars are still decades away

In My Humble Opinion
783

When a headlight is out, a cop pulls over your car and approaches the window to have a chat. But what happens when no one is driving the vehicle?

A video posted this weekend gives us some answers. On April 1, a Chevy Bolt operated by the autonomous driving company Cruise was pulled over in San Francisco (it’s unclear what infraction the robot violated). After the police officer peeked inside the car and saw that no one was in it, the car fled the scene…and parked again on the next block to await its fate.

In response to the video, Cruise said that the car followed the game plan: “Our AV yielded to the police vehicle, then pulled over to the nearest safe location…as intended.” Even better, the robot driver got away without a citation.

784
785

A couple days ago I observed a driverless car maneuvering around the pedestrian-choked streets near Old Town San Diego. With a chase van following right behind it and what looked like no driver at the wheel.

786

A self driving car that can be left at the entrance to a parking lot and later summoned when needed (an auto-valet feature) would be a major advance that would be greatly appreciated by drivers, especially those that are ‘spatially challenged’ in the parking skills department. But even this small subset of the driverless car technology seems to be rather slow to arrive.

Instead car makers seem to strive for grand all singing - all dancing solutions, that seem hopelessly ambitious.

I wonder why that might be?

787

Well, hubris, and a misunderstanding of the complexity of what they’re trying to achieve. Typical of that is Ford’s brand-new CEO, Jim Farley, who says:

If we can get people to fall asleep in our car, give them 45 minutes back on their commute, they can go to work 45 minutes later, they can go home 45 minutes earlier, it won’t be $500. It’ll be tens of thousands of dollars.

Ford CEO Jim Farley says customers will pay ‘tens of thousands of dollars’ to sleep in their cars | Fox Business

Maybe he’s just trying to demonstrate futuristic thinking, but nobody’s close to this, and I also think he’s wrong about what people will pay (if the thinks Ford customers and Tesla customers have the same levels of brand identification, he’s kidding himself).

788

A separate post for this NHTSA announcement, which has a hell of a finding in it.

NHTSA Upgrades Tesla Investigation Into Emergency Vehicle Crashes (cleantechnica.com)

The agency’s analysis of these sixteen subject first responder and road maintenance vehicle crashes indicated that Forward Collision Warnings (FCW) activated in the majority of incidents immediately prior to impact and that subsequent Automatic Emergency Braking (AEB) intervened in approximately half of the collisions. On average in these crashes, Autopilot aborted vehicle control less than one second prior to the first impact.

Musk has said several times that “autopilot wasn’t engaged when the crash occurred”, which appears to be true, but if it’s cutting out less than a second before the actual crash, I’d say it’s still a lie. I’m not putting all the fault on Tesla, of course; there’s also the drivers thinking their “autopilot” is true autopilot:

All subject crashes occurred on controlled-access highways. Where incident video was available, the approach to the first responder scene would have been visible to the driver an average of 8 seconds leading up to impact. Additional forensic data available for eleven of the collisions indicated that no drivers took evasive action between 2–5 seconds prior to impact, and the vehicle reported all had their hands on the steering wheel leading up to the impact. However, most drivers appeared to comply with the subject vehicle driver engagement system as evidenced by the hands-on wheel detection and nine of eleven vehicles exhibiting no driver engagement visual or chime alerts until the last minute preceding the collision (four of these exhibited no visual or chime alerts at all during the final Autopilot use cycle).

789

Auto-parking would absolutely be a game changer. It would be like everyone having their own personal valet with them all the time. Drive up to your destination and get out, and the car goes off and finds a parking spot. When you are ready to leave you call your car and it comes back from whatever spot it found and picks you up. I’d pay a large premium for that.

However, in some ways this is harder than just driving. There are a huge number of judgement calls involved in this. There’s also a lot of missing infrastructure needed to have the cars find open spots that are actually free and not closed by signage, figure out the price for parking, pay the parking fee, etc. What does the car do if there are no open spaces in the lot, and you’ve gone in for your meeting or whatever?

Also, parking lots have a lot of hazards. people walking out between cars, cars backing out without warning, etc.

We will probably need to see parking lots designed for autonomous cars, with broadcast status as to space availability, etc. The lots might be designed to mitigate some of the hazards. Parking meters will need digital payment.

But this is something we might see before full autonomous driving, and frankly I’d rather have that than the ability to sleep in the car. But if you have a 2-hour commute, YMMV.

790

If you are looking for a ride to store they might as well promise you a ride to the moon if they can’t even get you to the store.

791

How “large”, though? “Tens of thousands of dollars”, as Ford CEO Farley claimed? How many tens? On a Ford vehicle whose entire new-car price is in the “tens of thousands of dollars” range anyway?

I don’t think a lot of car buyers are going to be willing to pay the sort of premium for “auto-parking” capability that nearly rivals the annual wages of an actual valet/chauffeur. (Who can also work as a mechanic/handyperson/gardener/whatever when they’re not busy driving; let’s see an auto-parking system that can do that!)

792

I said nothing about tens of thousands of dollars. But for people that have to constantly drive and park, it would be a godsend. More so than sleeping in the car, which would only be valuable on very long commutes or long trips.

The ability to self-park would be used constantly. Just going to the grocery store would be easier.

793

I didn’t say you did, but you said that you’d “pay a large premium for that”. I not unreasonably wondered how large you considered an acceptable “large premium” to be in this case.

Especially given the recently referenced car company CEO’s estimate of an acceptable “large premium” for self-driving capability, which he placed at “tens of thousands of dollars”. (And as you yourself noted, there are aspects of auto-parking which are in fact harder than other forms of self-driving.)

794

An aside to the auto-parking discussion …

Around here (suburban greater Miami), metered parking on the street or in city-owned lots is commonplace. Especially in the older denser areas or near the beaches. Large commercial developments like hotels or shopping centers have the usual acres of free parking in attached lots.

Substantially every town around here has replaced the coin-op meters with smart meters attached to one of several phone apps. I know from these apps that SoFL is hardly the only part of the country / world where this is done, but it may be unfamiliar to many Dopers who live in less congested or newer surroundings.

With plentiful signage to tell you which app to use, the app will show you where the empty parking spaces are, turn-by-turn navigate you there, and enable you to pay all without leaving your car. As well you can buy, e.g., one hour of parking then if you’re running late, top up the “meter” for another 30 minutes or hour or whatever from your phone. And of course the app reminds you as your time is running low, so it’s hard to inadvertently overstay your paid time.

Except for the minor vexation of needing 4 apps by 4 separate vendors on your phone and having to learn each’s idiosyncrasies, it’s pretty magic. The 21st Century is really kinda cool. So far.

This capability is in no sense auto-parking, but you can see that it’s a big piece of the total required infrastructure for self-parking auto-valet cars. “All” we need now are cars that can see pedestrians and obstacles and drive themselves to follow the app’s directions to a chosen space. “All.”

795

I don’t think anyone will pay ‘tens of thousands of dollars’ for self driving - just like people didn’t pay Mercedes tens of thousands of dollars for air bags and abs. But what did happen is that the tech only appeared in cars that cost tens of thousands more than other cars - because they are luxury cars and can charge higher margins. That’s where all the new fancy tech first appears, then trickles down to the rest of us over time.

I could see BMW adding self-driving to the 7 series, while increasing the price. But eventually, the tech will make it into every vehicle if it works. Tesla might be the only company that could charge an extra $20,000 for real self driving, because, well, they’re Tesla.

What would I pay for ‘auto valet’ AI? Today, not much. I am semi-retired and seldom need to drive somewhere and find scarce parking. I am also relatively fit and don’t mind walking.

But there were times in the past when I would have paid a lot for it. Like when my kid was little, my wife worked downtown, we had lots of appoinments in office buildings without public parking, etc. Everyone’s use cases are different. Some people would kill for the service, others would be, ‘meh’.

796

Tesla has been shutting off autopilot just before accidents so they can weasel word their way into saying that people involved in crashes weren’t using autopilot at the time of the crash.

Investigation into Tesla crashes involving 'Autopilot' feature gets more serious | Fortune.

797

Pssst…like six or seven posts up. :slightly_smiling_face:

It will, but if ADAS is any indication, adoption will be pretty slow…it’s been available in many forms for a while, but a lot of people still don’t want to pay for it [and when it comes with the car & no optionality, they sometimes turn it off…but that’s a different discussion].

798

This is a tricky. I am not in any way excusing weasel words from Musk. If he says a car wasn’t on autopilot when the accident happened, because it was turned off a second before, then, yeah, he’s basically lying.

However, ignoring what Musk may say, there is nothing nefarious about the car turning off autopilot a second before the crash. When self driving gets in a situation that it can’t handle, it will turn off. That situation can be something as simple as the autopilot thinks it’s too close to another object. If the driver isn’t ready to take over immediately, there may be a collision.

I’ve had my full self driving turn off due to situations it got itself into. For example turning onto a street, and getting the front end too close to a parked car. No collision, because I’m paying attention, but the car freaked out. Self driving couldn’t have turned off or warned me earlier, because that parked car wasn’t too close earlier. Obviously, not getting too close to parked cars should be something it takes care of before it gets too close.

Other times it has gotten confused about the lanes, and turned off.

As for emergency vehicles, as far as a I can tell, Tesla’s full self driving doesn’t do a damn thing when it sees one. I recall the beta release notes mentioning something, but I’ve never seen any reaction in the car from flashing lights.

If nothing else, flashing lights that seem to be on the travel route, or approaching from behind should force the driver to take over. Better, obviously is to have the car do the correct thing, but until that’s ready, punt.

What I’d really like to see is that all emergency vehicles have transponders that publicly announce their current location and destination when their lights turn on. Then in car systems can monitor that, and know that there’s a fire truck approaching on that cross street, even when it’s out of view.

799

There are both nefarious and non-nefarious reasons that could happen.

800

Yeah, I don’t actually think it’s necessarily nefarious, but I do think he’s lying either way. That said, if it’s not nefarious, handing over control less than a second before impact is also a terrible peril recognition failure (which is essentially what NHTSA is already saying about Tesla’s recognition of emergency vehicles).

801

Absolutely, it’s definitely a failure of the self driving system to get in that situation in the first place. I consider it a failure in all of the unreported cases, where the driver took over before the system aborted, and so there never was an accident.

802

“True autopilot” is that there is still a (human) pilot in command that is responsible for the aircraft at all times. Autopilot is actually a great name for the Tesla system, in that it’s primarily a workload-reduction system, but does not actually replace the pilot. The issue, apparently, is that the public isn’t aware that aircraft autopilots don’t allow the pilots to take a nap while their aircraft navigate autonomously to the destination.