Self-Driving Cars

I haven't been posting a lot lately because there's not much in the way of specific news around self-driving, but reports from customers in Tesla's early beta of its complete Full Self-Driving package indicate that it is improving by leaps and bounds, with every new version allowing it to perform better and perform in more situations.

Elon Musk has been promising a wider beta rollout for some time, but I don't know the current status of that effort. In a recent informal interview with CleanTechnica, he revealed that Tesla's FSD team consists of some 300 engineers, split among various teams for components of hardware and software, and 500 labelers. He meets with them personally once a week.

Also, the long-awaited Tesla FSD subscription service may be coming out early next year... maybe (Teslarati article). It's been hinted at for some time. Elon continues to insist that purchasing it outright is a better long-term value, but for people who only lease their Teslas or don't intend to hold onto them long-term, a subscription model may make more sense.

Elon continues to go back and forth on the idea of an FSD purchase attaching to the customer rather than to the vehicle.

Anyway, the main reason I bumped is to carry something over from the general EV topic: Apple claims that it will enter the market in 2024 with a "revolutionary" battery technology and a lidar-based FSD solution. Little is known about this but it is intended to be a general-purpose package for everyday driving, not a dedicated system for a specialized task.

I'll be interested in what they come up with. Apple has a strong software focus and a lot of investment in AI and self-learning. If anyone can leverage institutional expertise (and loads of cash) into such a project, it can. However, Google (or rather its parent Alphabet) has been struggling to get Waymo, its own lidar-based solution, to deliver a general-purpose self-driving experience.

In precisely mapped, geo-fenced areas, lidar can be extremely accurate, but without those detailed 3D maps, it may struggle against a more general machine vision solution.

Edited by Fighteer on Dec 21st 2020 at 10:59:41 AM

It's always good to hear about possible advances in battery technology.

See, I'm intrigued about the Apple intervention in this market.

Only because Apple will have an advantage others will not: their brand and design philosophy gives an opportunity to redefine what a car looks like. And in so doing they might be able to unlock some of the sticking points we have when it comes to AVs.

As mentioned before in this thread, AVs perform the best in two areas: either highly controlled environments designed for cars such as Interstates or dense urban environments where you are lucky to average 10MPH.

Tesla appears to have the former sewn up, and as their self driving technology expands they are pushing the envelope in getting more interurban and rural roads covered by their technology. But in one aspect, Tesla are very traditional. They have built a private car based around how we currently drive: individuals operating independently. You own a Tesla: no one else is driving it (or being driven by it) but you.

It's the latter model of slow transport in urban areas where Apple needs to break the mold. This is where driverless cars would be closer to a form of public transport, replacing taxis and possibly busses. You wouldn't own the car, just pay a subscription to access the fleet of vehicles in the city. This is MAAS, or Mobility As A Service.

Because it's such a different approach, we need to revisit what a car even looks like. So throw out concepts such as a steering wheel. Don't have all the seats facing the direction of travel. It's telling that when I hear of this type of vehicle it's rare that they are called cars. Often they are referred to as pods. If you can think of a word more synergistic with Apple's brand than pod please let me know, I'll wait.

So, I'm waiting to see what Apple have planned. Their existing customer base means that if they do jump feet first into MAAS then they might have a big enough subscriber base to make it worthwhile. This really could be a revolutionary moment in transport. On the flip side however, Apple are also all about exclusivity, high price and status. It's also possible that their MAAS approach would just segregate public transport into the haves and have-nots, and I'm not sure that's a world we want to live in.

"Possible"? Unless Apple has had its entire corporate ethos replaced, make that "definitely".

Slightly off topic, but this debate reminds me of a "Why is this happening" podcast episode where the interviewee tried an experiment where she cut herself off as much as possible from each of the tech companies for a week: going so far as to run an IP blocker on her phone and computer and blocking the IPs from each provider directly.

Blocking Amazon was devastating because it also meant blocking AWS, and they practically host the internet. Even online shopping was impossible as independent shops and websites still use Amazon's fulfillment services. Blocking Google caused any geolocation service to crash, including Yelp ironically. Microsoft proved impossible, only because it finds itself in almost any smart device: cash registers, ATMs, your car entertainment system.

Blocking Apple? Surprisingly easy. Just put your i-phone down and you are practically free. It's because Apple has made a conscious decision to try and avoid monetising people's data. The profit margin on their hardware is ample enough thank you. So if you want your privacy, Apple are pretty good about it. But that's because they can: the premium you pay for their products allows them to provide that service. In effect, privacy and control over your data becomes something that only those who can afford Apple products can have. Privacy becomes a privilege of the rich.

It's something likely to happen here if Apple takes that approach with MAAS. Think of public transport. Think of the stereotype of the homeless person sleeping on an Underground train. You wouldn't get that in an Apple pod. It might not all be bad: Apple may result in private vehicles being taken off the street as ride sharing becomes commonplace, which would reduce congestion and improve bus reliability. But in reality if you pay for an exclusive MAAS package from Apple, you ain't going to be paying public transport fares. It will be up to governments to either make up the difference, or manage the decline in services. You don't have to travel too far down the slippery slope before you see a two tier public transport system with Apple providing shiny high quality services to those who can afford it, and everyone else suffering with an old and decrepit service that will never get the investment it needs.

Before we get too deep into that rabbit hole, I must point out that Tesla's ultimate goal is to supplant private vehicle ownership as well, or at least make it so that most people don't need to own a car because self-driving robo-taxis are cheap enough and ubiquitous enough for even the poorest person to have access.

Individuals who own cars would be able to second them to the robo-taxi network and earn income by doing so, but that reaches a point of maximum efficiency and then has diminishing returns. Anyway, most cars would be owned and operated by large fleets that also provide the garaging, charging, and maintenance.

If Apple is also working towards a fleet robo-taxi approach, then its goals and Tesla's will converge, even if Apple is starting from a point farther down that road. I'm not really concerned about Apple going for the boutique/high end robo-taxi club, because it will be entering a market that already has an innovative competitor. It won't be able to parlay first-mover advantage into a de facto monopoly.

In that sense, having two players each taking independent approaches will be a net positive for consumers, as single-player markets tend inevitably towards monopolistic behavior no matter how well-intentioned they start out as.

Personally I find this ironic as Tesla is often called the "Apple of EVs" for building out its own proprietary software and hardware environment. Now that Apple itself is entering that market, it kind of breaks the metaphor.

On the subject of mass transit, there is already a problem with battery-electric vehicles and transport in general: most road building/maintenance is paid for with taxes on gasoline/petrol, and of course EV owners end up not paying those. There are already moves by some jurisdictions to tax them directly to deal with that problem.

If robo-taxis divert funding from mass transit, a similar tax would seem to be in order. Of course, there's always the possibility that they do the job well enough that we don't need buses and trains, although I'm skeptical.

Edited by Fighteer on Dec 22nd 2020 at 7:42:40 AM

Yeah, I don't think that (autotaxis replacing buses and trains) would be desirable. Per capita power (and space, in some circumstances ; ) consumption would increase and thus the environmental impacts.

The talk about Apple and data sales does raise a concern about robo-taxis and self-driving cars where the software is owned by a tech company, someone deciding to have the car sell your location data to advertisers as the default. So if you have the car take you to work every day you will start getting targeted advertisements for businesses by your workplace.

Now if people consciously choose to sell their data that’s fair enough, but it’d be very on-brand for a tech company to just start selling your location data as standard without mentioning it.

It was pointed out that Apple doesn't need to sell your data because it already collects a huge premium on the hardware sales as well as a cut from the App Store. Of course, those apps are free to market your data to pay for themselves. For example, your smartphone tells Facebook where you are whenever you open the app, and Facebook will happily, nay eagerly sell that information to advertisers. (Yes, you can disable that, I'm making a general point.)

Privacy issues aside, having location-based data trigger recommendations is generally positive for consumers. If I'm heading out for lunch, I probably have an interest in finding out if there are any special deals nearby. As long as I can turn that feature off if I want to, I don't see the problem.

Anyway, revenue from self-driving cars will generally come in three areas:

• The original sale of the hardware (akin to the phone). This may of course be leased or paid in installments, whatever.
• A cut of the fees for robo-taxi service.
• Premium infotainment services. This is the new market that businesses are going to start wanting a piece of. As cars no longer need to make sure that the driver is paying attention to the road, the opportunities for infotainment are going to skyrocket and potentially pay off big.

Think, essentially, of the App Store or Google Play store or whatever expanding to cars. Tesla is already working on this: rolling its own, as it were. Apple of course is the industry leader in app stores, so we should expect its cars to focus heavily on the concept.

ETA: While I am skeptical that Apple can produce enough cars fast enough to catch up to Tesla in market penetration, especially if it's partnering rather than vertically integrating, I am happy that we are getting more options. Consumer choice is a general positive, and I don't want Tesla monopolizing self-driving solely because everybody else gives up and goes home.

That said, I have a hard time picturing the kind of person who would reject a Tesla car because they hate Elon Musk or whatever but happily get into an Apple car... presumably based on their deep and abiding love of Tim Cook.

Edited by Fighteer on Dec 22nd 2020 at 12:59:58 PM

Teslarati: Tesla FSD Beta performs flawless U-turn in real-world inner-city test

The linked video shows a Tesla Model 3 performing a U-turn in Long Beach, CA without any intervention from the human driver. I've seen some other impressive videos lately, such as left turns at uncontrolled intersections. The beta software is getting better in leaps and bounds.

Elon wasn't kidding when he said that the new architecture would lead to rapid improvement versus the old one.

Edited by Fighteer on Dec 28th 2020 at 11:29:16 AM

In this Business Insider interview with Elon Musk, he states that he firmly believes Tesla vehicles will be capable of Level 5 full self-driving by the end of 2021. He gives this a "100% certainty", which is pretty ambitious, with the major hurdle being to convince regulators that the cars are safe. I'm withholding judgment, since he's been off on the FSD calls in the past.

When asked about the process of transitioning from human-driven to autonomous cars, he has a lot to say. First, the most dangerous period will be Level 4, when the cars are good enough to drive themselves everywhere but still need humans behind the wheel. At that point, they'll be able to handle 99.9 percent of road situations, but there will always be that 0.1 percent or 0.01 percent that fools them, and the problem will become one of "chasing the nines" and rooting out the edge cases.

Musk thinks that by the time self-driving cars are ten times safer than a human, it will start becoming harder to get a driver's license. Right now there's enormous pressure to hand them out to anyone who can get behind a wheel and not run into a building (yes, I know rules are stricter in certain countries) because they're economically necessary. That will no longer be the case. Humans and self-driving cars on the road together will be inherently dangerous until we replace the humans (my interpretation, not his exact words).

He doesn't believe that manual driving will ever go away completely, but the requirements for being licensed will go up and it'll become a luxury/sport thing rather than a necessity. "I'm definitely not trying to take anyone's steering wheel away from them." This was in response to a past statement that the interviewer brought up: "I don't care about the moment when self-driving will be permitted. I care about the moment when it will be forbidden for human beings to drive a car."

For Tesla to replace one percent of the global fleet per year, it would need to manufacture 20 million vehicles, every single one of which is equipped for complete autonomy. That is the goal.

Edited by Fighteer on Dec 30th 2020 at 1:49:45 PM

Ars Technica: Mobileye CEO dismisses Tesla self-driving beta as a “crappy system”

Click bait title from Ars Technica, but underneath is a well-researched article discussing Mobileye's approach to self-driving and how it contrasts to Tesla's. Currently, Mobileye (owned by Intel) licenses advanced driver assistance system (ADAS) technology to OEMs and collects data from the vehicles they sell. Tesla, in turn, manufacturers its own ADAS and deploys it exclusively on its own vehicles.

Both use machine vision and eschew lidar, although Mobileye intends to deploy a lidar solution in parallel with its machine vision solution for increased reliability. Tesla relies on image processing to recognize and react to its environment in real time; Mobileye builds high-definition maps and constructs databases of observed driver behavior to allow its vehicles to behave properly.

It will be an interesting race to see who wins, but we could all benefit from the competition.

From Twitter, Tesla Owners of Silicon Valley posts a clip showing how Tesla FSD Beta safely navigated a seriously tricky situation: passing a delivery truck on a two-lane road with oncoming traffic. The full video is here.

It's worth watching in its entirely because damn this software is getting good.

The passenger, @Real_Futurist, tweeted about his experience - first time with FSD (edited for format):

Edited by Fighteer on Feb 4th 2021 at 8:08:19 AM

I'm watching that thing and naturally comparing what I would have done in the same situation, and I'm honestly not sure which of us (the computer or me) is the better driver.

The fact that we are even asking the question should be proof that they've done something amazing.

Elon keeps getting asked when they'll release FSD Beta to the wider public and he says that they have a serious human behavior problem to contend with. When a system like this is good in many situations but still dopey in enough of them, people pay active attention. But when it's really good: flawless in all but a few edge cases, attention wavers and people may miss critical interventions.

FSD is coming, but it can't just be 99.9 percent reliable. It has to be 99.99 or 99.999 percent to achieve widespread adoption.

Cadillac (a GM brand) posted a Super Bowl commercial for its "hands-free" driving platform titled SuperCruise, available on the upcoming 2023 Lyriq and a few other models. The commercial is actually fairly clever, with Timothée Chalamet as Edward Scissorhands' son, Edgar, having difficulty with many tasks until he lets the Cadillac drive for him. And yes, that is Winona Ryder.

(For a brief moment I wondered why Paul Atreides was driving a car.)

GM would like everyone to believe that this is a revolutionary new technology but it is limited to a specific set of roads that have been precisely mapped. Obviously, it is under no obligation to advertise for a competitor but it's hard to see how anyone should be expected to take SuperCruise seriously compared to Tesla Autopilot.

Edited by Fighteer on Feb 8th 2021 at 6:46:38 AM

I came here to post "wait, hands free? Tesla Autopilot isn't hands free, isn't that apples to oranges?" And indeed, that is the case, according to Tesla's own Autopilot page where it says "Autopilot and Full Self-Driving Capability are intended for use with a fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment". (I still take issue with the "Full Self-Driving" branding, but that's an aside.)

But looking into it, it turns out that GM's Super Cruise feature has been out since 2017 and already allows hands-free driving on mapped roads. The amount of roads that have been mapped is increasing every year, though earlier versions of Super Cruise have to get a software update at a dealership to use the newly-mapped roads. I am unclear if this is an update to the self-driving system itself or just uploading the new maps.

So Super Cruise can do things right now, and in fact has been doing things for three years, that Autopilot cannot yet do. Yes, it relies on mapped roads, and there are arguments that Autopilot is (or will eventually be) the superior system. But Super Cruise is still doing things that Autopilot can't, which is a far cry from "it's hard to see how anyone should be expected to take Super Cruise seriously compared to Tesla Autopilot."

Edited by NativeJovian on Feb 9th 2021 at 9:55:49 AM

Having played video games ever in my life, I can confidently say that an assuming someone in a self-driving car will be 100% ready to take over and avoid an accident at a moment's notice is pure fantasy from the beginning.

Edited by Clarste on Feb 9th 2021 at 8:01:16 AM

Can someone clarify, does the Super Cruise feature allows hands free from a technological perspective or a legal perspective?

Because I think Tesla would argue they’ve got hands free driving ready to go from a technological perspective, it’s the law that require a driver to be paying attention and have their hands on the wheel.

So that’s the question on Super Cruise, if if activate the system on a mapped road and let it drive, with you doing something else, are you breaking the law?

The big barrier to self-driving right now seems to be legality, a government somewhere has to sign off on you being able to delegate the driving to the car itself. Which I expect will be done on specialised roads first, though didn’t we have some news a bit ago about a city in the US doing a test rollout? Plus isn’t Tesla doing something with Musk’s tunnelling company where you’d use self-driving in a road tunnel they’re making? As the tunnel would presumably be a private road and subject to owner choice rather than public road rules (same as you can let a kid drive a car on private land).

Come on. Absent pathology or intoxication, the human brain is able to recognize the difference between driving in a video game and driving in real life. Yes, it's human nature to pay less attention once the car starts doing some of the work for you, but we can't reach self-driving if we don't have humans in the decision loop in the early stages.

The measure of success in self-driving seems to be our willingness to tolerate the learning curve when it won't be as good as a human in all situations. Europe, for example, is taking the "bubble wrap" approach where self-driving is prohibited until it's super safe, which means that it'll take a lot longer to develop without the opportunity for large-scale real-world testing.

Meanwhile, the U.S. could be said to be taking the "cowboy" approach where it's a free-for-all, damn the consequences. Could be said, but isn't, because we all know what happened to Uber. A single fatality led to the effective cancellation of its program.

Tesla gets around this by putting in explicit language that the driver must be in control at all times. It can also permanently disable FSD for drivers who use it irresponsibly.

I don't know the legal framework around GM SuperCruise. It's probably worth looking into. As it stands, it is only capable of operating on specific highways, and those are much less dynamic and less hazardous overall than city driving. Sure, the speeds are faster, but you don't have to deal with intersections and everyone's basically going the same direction.

If Edgar Scissorhands only drives on highways, he's probably fine, but as soon as he has to take an offramp, his disability will start to be a problem. Ironically, if he were in a Tesla, the ad could legitimately make a claim that it could take him from door to door without interventions.

I think we need to make a distinction between driver assist systems that are Level 4 but only in very specific situations and ones that are Level 2 or Level 3 but can handle almost anything. The problem with GM's approach (and Waymo's) is that it's very difficult to scale. Great, it can handle ~130,000 miles of roads. There are over 4 million miles of roads in the U.S. SuperCruise Tesla Autopilot can in principle handle 99 percent of them. SuperCruise can handle about 3 percent.

130,000 miles of mapped roads was the number as of the initial launch of the program in 2017. As of the beginning of 2020, it was over 200,000.

Also, GM is working on "Ultra Cruise", which will work on city streets rather than just highways.

Just for comparison, Consumer reports rated 17 different driver assistance systems in 2020 and put Super Cruise at the top, with Autopilot a "distant second". Autopilot got slightly better scores on "capabilities and performance" and considerably better ease of use, but Super Cruise was rated significantly higher for monitoring driver engagement, making it clear when it was safe to use the system in the first place, and dealing with an unresponsive driver. Those things may not be as sexy as actual ability to control the vehicle, but for something ultimately intended to increase vehicle safety, they're pretty damned important. Overall, they gave Super Cruise a rating of 69, while Autopilot got a 57, and most of the rest of the systems were in the 40s.

So yeah. The idea that Autopilot is head and shoulders superior and everything else on the market is just a sideshow doesn't really reflect reality. You can contest the opinion that Super Cruise is better if you want (and you may well be better off with Autopilot, depending on your particular use case), but it's definitely not true that Autopilot is obviously and significantly superior in all cases.

Fair enough. I'm very curious to see who wins the race in the end.

Clearly, we must have a demolition derby to determine the winner.

To make the point, the winner would be the one that doesn't crash.

~Silasw: I'm moving the conversation to the proper thread.

First, consumption of marijuana, however pleasurable or medically useful it may be, is not a fundamental requirement for society to operate. Transportation is. (I've never used cannabis, so I can't offer a direct opinion on its properties, but the fact that I have lived this long without doing so tells me that it is not a baseline necessity. Your cousin who tells you, "I need my weed, man," is an addict and you should not get economic analysis from them.)

Second, while there will undoubtedly be intense resistance, the adoption of robo-taxis will be driven by sheer economic value: it's cheaper to operate a robo-taxi than a human-driven taxi. Self-driving will similarly carry the benefit of eliminating auto insurance, never mind the trillions of dollars of economic losses from personal injury, death, accident cleanup, road maintenance caused by accidents, auto repair... the list goes on and on.

Robo-taxis gain more utilization from the vehicle itself. Most cars are driven only a few hours a day, if that, and spend the rest of their time in parking lots or garages. If we can increase that utilization even to fifty percent (12 hours a day), we can reduce the total number of vehicles required to transport people by at least half, and that alone will lead to less environmental damage (regardless of power source), better land utilization, and many other benefits.

If people can get a robo-taxi whenever they need it and the aggregate cost is lower than owning a car, the argument for personal car ownership will disappear almost entirely, save for people who do actually use their vehicles for large parts of a day. This in turn means that garages and driveways won't be necessary for most dwellings.

If most people are not driving their cars, then time spent in cars becomes more productive and/or it becomes possible to offer value-added services like in-car entertainment that are simply not practical today. This opens entirely new market opportunities.

If cars last for millions of miles rather than a few hundred thousand and aren't constantly getting wrecked in crashes, then less resources need to be consumed to build them, further reducing overall consumption with economic and environmental benefits.

The argument for self-driving robo-taxis is so overwhelming that it's not a question of if but when. To be sure, we'll eventually reach a peak of market saturation that drives prices down until it's only barely profitable to operate them, but that's a general economic principle that applies everywhere... or rather it should, but we've seen that it doesn't hold true in practice because monopoly power always arises to prop up profit margins. That's an entirely separate argument.

Anyway, it'll be a long time before that saturation point is reached and there is an opportunity for a first-mover to reap huge profits in the meantime.

I'm going to edit in another point instead of double-posting, since this hasn't gotten any replies yet.

I hear an argument being made that giving up personal car ownership is anti-democratic; that it's surrendering to "corporate control" or something like that. This argument is, ironically, one that big business loves because they created it themselves.

Yes, the idea that the American model of personal freedom requires individual car ownership is one designed, marketed, and maintained by Big Auto. Why? So they can sell you the latest Big Truck 5000 with custom penis extension... or the latest Family Megavan that lets you sit above all those horrible men trying to steal your kids.

"You can drive anywhere..." as long as you buy gasoline from Big Oil. "You can own your own car..." as long as you pay interest to Big Finance. "You can have personal independence..." as long as you pay thousands of dollars to auto dealers for service on your car that's designed to break down frequently. "You can go from 0-100 in 0.8 seconds..." as long as you carry insurance to cover the risk that your reckless driving will land you or others in the morgue.

Every aspect of this ideology is carefully crafted to trap you in a cycle of consumption. Self-driving is freedom from it and the people who stand to lose are well aware of that fact. Consider the sources of your skepticism before you indulge it.

Edited by Fighteer on Feb 26th 2021 at 9:20:35 AM