Who Will Win the Driverless Car War?
Image by Doug Kline 
[Content note: This post is wholly speculative. I don’t own shares in any of the companies mentioned.]
What’s happening with cars?
What do you mean?
Everyone is trying to make one!
Oh, that. Well, you do know that in the early years there were literally hundreds of car manufacturers based in the US, and now there are two main ones?
I’m not talking about the boring companies.
Oh, that’s right. Apple is trying to make cars. Google Alphabet is trying to make cars. Uber is trying to make cars. Tesla is trying to make them. And now Samsung too, don’t forget about Lyft (also before they get replaced remember some tips for Lyft drivers which might come in handy).
Carnegie Mellon had some interesting stuff going on too until they entered into a “partnership” with Uber that resulted in Uber hiring away all of CMU’s relevant researchers.
Wait. Tesla already makes cars.
They do, But they are also working on a car for the next sea change, which they believe to be autonomous driving.
What is actually going to be the next “sea change”?
Autonomous driving. Another possibility is that it will be autonomous driving along with a change in who owns vehicles. That’s what Uber is hoping for.
What about just having a nice infotainment system? Wouldn’t that itself be enough?
I think it’s reasonable to consider a “sea change” to have occurred when a significant number of existing companies in the industry aren’t able to survive the change.
It’s possible that Apple doesn’t care about autonomous driving and really just wants to build a car with a nice infotainment system. I don’t think that’s likely though, and it probably wouldn’t result in what I am defining as a “sea change”. If you want a car with a usable infotainment system, you can get one today. Buy a 2016 Honda Accord. It outsources everything to your smartphone via Apple CarPlay or Android Auto.
What about technical problems?
I keep hearing on Twitter that autonomous driving will be here in 15 years. But you should generally be dubious of any estimate someone in IT gives you. Quintuple this doubt if the unit of measure provided is years. I think the 15-year estimate is at least a tacit admission that this is a really hard problem. And it isn’t totally clear that everyone has considered every dimension of the problem.
Can these other companies actually build a car?
Yes. There are plenty of suppliers who will be willing to help them. There are multiple suppliers who will provide you with a full drivetrain, an interior, brake systems, exhaust systems, etc. They could probably be easily persuaded into putting the whole thing together for you too.
The real question is at what price can a Silicon Valley company build a car. The auto industry doesn’t have many new entrants in large part because the way to success (for anyone who isn’t a specialty manufacturer) is market share. Developing a platform costs billions. Building a plant and buying machinery for that platform costs billions. The only way to get that investment to pay off is to sell a ton of cars and spread those costs across all of them.
Building a car isn’t an impossible engineering task. Building a new nameplate that can sell in volume almost immediately is a Herculean business task. But it’s one that Google, Uber, Tesla, or Apple could probably pull off if they ever get around to developing the right product.
What is the right product?
The dream is a car that drives safely unattended.
Who’s in the best position to solve this problem?
To answer that question, we need to know what are the fundamental skills most important to building the next generation car.
If mapping ends up being the most important thing, Google would hold the advantage. Additionally, Google has been running its Prii around California for a good number of miles.
Uber owns an important network that anyone who could possibly be considered an early adopter is already familiar with. If Uber were to deploy autonomous cars, you as a consumer wouldn’t have to ask yourself if you want to spend $70,000 to buy an autonomous vehicle. You would just hire an Uber at minimal outlay. Uber could profitably build $250,000 cars for quite a while without needing to worry about selling them to anyone else. Another advantage Uber has over its competitors is that they have a built-in solution for the problem of who absorbs liability for accidents. They would be both the manufacturer and the consumer and could self-insure their vehicles in case of an accident.
Tesla is already making cars and doing a good job of it. Additionally, their current cars have a bunch of near-autonomous technologies. For Tesla, it’s just about removing the last 1% of situations their software isn’t yet able to handle. How long will that take? Elon Musk says two years.
Apple…has a lot of money. And they seem to have some sort of knack for doing things properly after others have done them poorly.
Many existing car makers also have autonomous vehicle programs. If solving autonomous driving ends up being an easy problem, then these companies will have a great cost advantage having had decades of experience competing on price in the industry.
That wasn’t actually an answer.
Tesla. They are cool enough to hire first-in-class programmers, but they also have tons of data from real drivers. And not just drivers puttering around in perfect weather in California. For a problem as tough as autonomous driving, their approach of paring away that last 1% is the right one.
My second pick would be Google since they are at least puttering around in driverless cars on real roads now. Which drive like an autism stereotype.
What about that guy who built a prototype in a month?
George Hotz? I’m impressed that he did it in a month. And I think it’s really cool and want to be his friend. But there is a huge difference between making a working demo that will work 99% of the time and figuring out how to complete that last 1%. In particular, he is teaching his car to drive like he drives. Unless he finds a way to expose himself to crash conditions, the system won’t have data on what to do in those situations.
His next step probably needs to be creating a simulator so he can do just that–repeatedly expose himself to bad behavior by other drivers. His idea of just collecting more data by driving for Uber won’t necessarily give him the requisite variety.
Even after such training, he needs to figure out how his system will perform in situations he hasn’t trained it for. There will probably always be unanticipated situations that lie outside of the system’s training set. How it generalize to unfamiliar situations? If a truck full of bowling balls drops its load on the freeway, you know what to do even if you’ve never thought about it before. What would Hotz’s car do?
Hotz needs more people, but he comes across as a person who isn’t interested in or capable of running an organization.
But maybe he or someone else like him can. I’d be happy to see a company not mentioned in this post figure it out.
But at least some of that applies to large auto manufacturers today.
The ability to hire programmers does not. General Motors and Ford invested millions and millions and millions into Ford Sync, MyTouch, and GM OnStar. They wanted to do it themselves, but in the end this just meant hiring consulting firms to do the work while they retained IP rights. These ended up being terrible, buggy systems, and I feel bad for anyone who has used them. It isn’t a question of funding or will. Autonomous driving lies outside the capability sets of these companies.
That’s not the only big, ambitious investment GM has made. They attempted to make a revolutionary car that would transcend their existing businesses. They gave the new team unprecedented independence and generous resources. The result was the Chevy Volt. This car did, in fact, end up saving the company through positive PR that plugged into the limitless optimism of Americans in believing they can save the environment by buying more stuff, but the world still seems unimpressed with the car itself.
If the large auto makers survive, it will be because they are able to license autonomous driving systems from one or more suppliers. If Google gets there first, that might be exactly what they do.
What happens when someone wins?
This will depend on who gets there first and what the solution ends up being. If it’s Google, they will probably just end up licensing their software to existing car manufacturers. The makers will unveil high-end models that allow for autonomous driving. This isn’t all that different than what would happen if the car manufacturers were to figure it out themselves.
If Tesla or Apple win, they will introduce their own vehicles that will compete alongside everyone else’s. Tesla might be willing to license their technology to other manufacturers. Apple wouldn’t.
If Uber wins, they will build cars for themselves and deploy them in various cities, not necessarily in the US first. I suspect that if they think there would be sufficient demand, Uber would eventually be willing to sell autonomous cars to the public.
Of course, there can be more than one winner. For whatever reason, Apple’s Siri was followed closely by Google Now and Microsoft Cortana. The tech industry may be incestuous enough that one company’s product will be closely followed by another’s. We’d then see multiple autonomous offerings competing against each other.
Is there a question you would like me to ask now?
Yes. On which dimensions will these companies compete?
The most important thing initially will be whether it is actually more convenient than driving yourself. Autonomous systems may give cyclists full use of their lane even if they are riding in the shoulder. Will a driverless car then potentially get stuck behind a bicyclist for 5 minutes?
I know 5 minutes doesn’t sound like a long time, but it would feel like it to a rider who perceives he could do a better job.
There are a million such real-world annoyances. A less-annoying vehicle will outsell a more annoying one. This will be more important than price for at least their first decade. People will read reviews to figure out whether a car’s autonomous system will do a good job on their particular commute. If Apple is to play a part in this future, it will likely be because they seem to have a knack for making annoying types of products somewhat less annoying than their competitors’ versions. Uber might also be a beneficiary. My guess is that people would be really hesitant to buy an annoying autonomous system, but more willing to take an Uber ride with one.
Will the lawyers actually let any of this happen? I keep hearing about legal liabilities being the biggest impediment.
These companies we’ve been talking about own the known universe. If they want the laws to change, they will change.
I see the Google cars (the Lexus and occasionally those tiny, funny looking ones with a face) in the same places, at the same times of day, every day now. I assume they just follow the same routes: through neighborhoods, around parking lots, down major surface arteries, under drive-time and off-hour conditions. I’ve only seen one on the highway once, and I have no way of knowing whether it was self-driving or being driven manually, but it was in the right lane and doing 55 in a 65, and people were going around it.
My theory is that the first non-test driverless cars will be either Uber or a Google version of Uber (because Google can create a version of pretty much whatever it wants, at this point), because they will get around the obvious and difficult legal and liability issues that way (a point you make, of course), and Google seems to be figuring out most of the non-highway driving issues. And I bet we see them before 2025, though I have no idea when we’ll start seeing people riding in their own driverless vehicles.Report
My father was a field auditor and safety engineer for most of twenty years, putting on >50K miles per year, and never had an accident. At the first opportunity after I got my license, he made me chauffeur him around for a week to, as he said, “put some polish on.” One of the points he pounded at me all that week was that driving in traffic, particularly heavy traffic, is a herd endeavor, not a solitary one. Good self-driving software is going to have to be able to handle that, until we get to the day when most of the cars are autonomous.Report
The goosinator can already play the chase game, and in variable environments too.
Follow large and well-identified object is a game computers play well.
Deal with Ten Pineapples bouncing down the road? Less well.
Deal with “possibility of uhaul depositing crap on road in the next 10 miles”? Probably pretty poorly. (correct answer is always pass the uhaul).Report
Yes but what will the dronecar do if it’s following that truck that spilled the pineapples on a double yellow line? Answer: not pass or go around. That would be illegal. It’d probably just stop in the middle of the road.Report
Dealing with actual “avoid this” situations ought to engage minimal safety protocols (which include breaking the law in minor ways). In this case, you want it to drive on the shoulder, not cross the yellow line. (Is driving on the shoulder illegal? It probably is for long distances, under the heading “unsafe driver, possibly impaired”)Report
What if there is no shoulder? Or the shoulder a jersey barrier? What if there are two lanes over a bridge and the side you’re on is blocked?
How do you define “breaking the law in a minor ways”? 5 mph over the speed limit? Passing on a double line? Failure to come to a complete stop on Red?
Oh, and here’s a good one. In my state it is illegal to NOT move left if an emergency vehicle is on the shoulder, ie you’re in the right lane and the EV is on the shoulder, you are required to move to the middle lane. Now, take that scenario to an off ramp. The EV is on the shoulder of the off ramp, There is only one lane on the off ramp. It’s technically illegal to use the off ramp! And yes, I’ve been in court when a guy who was cited for doing just this was cited by a cop. His fine was 500 dollars. How exactly will a driver less car conduct itself? There is no shoulder, it’s blocked by the EV. It cannot move over the required amount, there is no more road. Will it stop and wait?Report
To your last question, I’m assuming, by the commentary seen above about Google cars, that the answer for them is “stop and wait” — because this is PA, and the quickest way to piss off the drivers behind you is to actually stop for the stop sign on the onramp to an interstate. If they’re reporting that “following the law” is causing more accidents, that’s probably what they’re doing “wrong”.
To the initial question: Occasionally, the best answer is really to brake hard. Dodging the pineapples or simply going over them are also good options.
I think once there’s credible math behind the driverless cars, we can get away from them following every law to the letter. Imagine: “we ran 100 simulations of the road (and the current drivers on said road), and this was the safest speed.” But you have to be able to get lawyers to sign off on that, and actuaries (who I have no reason to doubt will jump at the chance).Report
I’m sure all problems can be solved, it’s just how. I’m, at this point, not sure how you wrote code that you know, knowing will instruct the car to violate the law, and not end up getting sued for it when the car does, and that act is the proximate cause for some injury or damage. I think, for the time being, the code will be to “stay within the law” which will make driver less cars more of an annoyance to non driver less cars.Report
I think you’re righter than I am. Change the law so that there are fewer impediments to safe driving. Luckily, uber has experience at law-changing.Report
Assuming autonomous cars become the norm, I expect that the rate of accidents caused by insufficiently cautious humans rear-ending over cautious robots will decline significantly, to the point where robot cars will very soon have much lower accident rates than human driven ones.
That is for the same reason that the number one thing that makes cycling safer, far ahead of every other factor, is cycling – the more people riding bikes, and the longer that’s been the case, the more everyone is used to people riding bikes, expects them, looks out for them, accounts for them in their defensive driving strategies.
We deal poorly with the unexpected. And it seems that right now an (arguably) appropriate degree of caution about not running down pedestrians, is unexpected to a lot of people on the road. If robot cars reach 5% of the vehicles on the road, that will not be unexpected. And that 5% robot vehicles will probably be enough to slow down the overall speed of traffic enough to significantly reduce not only their own accident rate, but the rate of accidents not involving robots.Report
Well on the one hand if the autonomous cars speak to each other this will make them potentially extremely effecient and safe in a herd. If more than one producer makes them some common standard will be required or that assumption goes away.
On the other hand if the cars speak to each other then the potential for hackers to cause enormous hell becomes immense.
On the other other hand, if the cars speak to each other and have adequate protection from hackers then suddenly non-autonomous cars will become the least safe vehicles on the road.Report
Except for those vehicles that don’t aren’t hack able. Kinda hard to hack a car with out a computer module.Report
You can have car to car communication without the whole thing being remotely hackable*.
Problem is, once autonomous, first thing the police will want is a way to control a given car remotely, hence they will be hackable.
*If a hacker has physical access to a vehicle, all bets are off.Report
Actually, you won’t need to “hack” it. It’ll be code that is required in the OS by FEDGOV.
But I was really talking about cars with engines that predate much electronics. I’m not sure how you “hack” a car with a carb and no computer module.Report
I’m not sure how you “hack” a car with a carb and no computer module.
Brute force.Report
With a hacksaw.
like this guy.Report
By which you mean 1950’s cars and trucks. Anything past about 1970’s has enough electronics to interfere with radio instrumentation.Report
I was thinking more along the lines of the simple “herd behavior” rules that get used in animation these days. IIRC, there are two parameters in that model and by adjusting them you get all of running herd of herbivores, swimming school of fish, and flying flock of birds. Given distance sensors, it ought to be straightforward (I avoid “simple”) to have rules that, if followed by each vehicle individually, give rise to cooperative behavior.Report
Emergent behavior is going to be interesting…
But that’s why they have game designers working on the systems.Report
Someone already hacked the googlecar into doing doughnuts in the parkinglot. They fixed that security hole the next upgrade.Report
What is actually going to be the next “sea change”?
Amphibious cars.Report
Related:
http://www.smbc-comics.com/index.php?id=3692Report
That’s a good, solid summary. I think the Leaf means more than you think it means. I think it’s a slow burn, though. In many ways, it’s kind of a niche product, a vehicle that’s really only suitable as a second car, that you use to go to work and on short errands. But at that task, it does really, really well, and pleases the people I know who own it to no end. But no, it’s not autonomous.
And yeah, organizations who don’t do software often struggle mightily when they try to add software to their capability. I think it’s a problem of “how hard can it be?”Report
My friend the software engineer says that it’s mostly a legal issue.
We already have autonomous trucks on limited access roads, after all.
Uber has the lead, not because of the CMU profs they stole (total PR operation, that), but because they’ve got more experience at crashing through legal red tape.
And you forgot a question: How easy will they be to hack?
[disclaimer: I may benefit financially from some of the above discussed companies. Don’t ask me which ones, I really don’t know].Report
Sea changes are really rare. Incremental progress is the usual thing. In two years from now I could easily see a car being sold with a hands-off freeway mode, especially if the California legislature and the big insurers are willing to agree. (But will autonomous cars exceed the speed limit? The safe speed on the 5 freeway is often higher than the posted speed limit.) Between lane control and collision prevention, cars are a lot of the way there already. The additional feature would simply be telling the car what exit you want. The city street version would be next. Country roads, snowy conditions and other predictably challenging environments would be last. And for a generation there will always be a manual over-ride.
Early versions are bound to be rocky. I imagine it would be more tiring, not less, to be behind the wheel of an ordinary car that just so happens to handle the freeway driving. But since the benefits appear so substantial, I think that car makers, legislators and insurers will be willing to press forward.Report
What’s even rarer is for everyone to announce years in advance “This sea change will be happening shortly” and then have it actually happen. People tend to be famously lousy futurists, but people have a stunning amount of confidence around this particular idea.Report
UBER is betatesting the necessary problem-solving — how to steamroll governments.
You can assume that we’re basically done with the coding. (we aren’t, but the designs are all written).Report
@francis is correct. Truly autonomous cars will be an evolutionary process and the end result will bear little resemblance to cars today. The biggest driver of the time frame for this evolutionary process is the product lifespan of cars.Report
Interesting and topical. A piece by Cory Doctorow discussing the problem of coding to deal with odd ethically challenging situations and also protecting that code.
http://www.theguardian.com/technology/2015/dec/23/the-problem-with-self-driving-cars-who-controls-the-codeReport
As someone who would not be able to tell this himself, I wonder if someone out there who has his or her finger on the pulse of the whatever you would call it would be able to say whether there has been a truly *SEXY* autonomous car released yet.
I don’t even know what kind of example to give. My first intuition is to say “You know. Like one of those firebirds that had the art on the hood of the car” but there’s a little voice inside my head saying that those haven’t been sexy for a while.
But is there one of those that is autonomous yet? Because that seems like it’d be an important milestone.Report
What’s the sexiest car design ever?
I’m not sure if it’s my favorite design (though it’s pretty great); but the 3rd-Gen Corvette Stingrays (1968-1983) – you know, the ones where they just said “screw it, let’s shape this thing like a woman” – have to take the cake.Report
There’s a yellow 1972 Corvette that I pass every day on the bus, and I frequently think, “Man, at one time cars were an artform.”Report
I’ve never driven one, so I have no idea what it’s like to sit so far back and have such a long hood; but DAMN those things are beautiful.Report
For masochists, the Pontiac Aztek. For exhibitionists, the AMC Pacer.Report
How much does it cost to get a good reliable Hummer? wink wink nudge nudgeReport
And for pyromaniacs, the Ford Pinto.Report
If Tesla produces one, it will almost certainly be sexy.Report
Frankly, I don’t find the Tesla S THAT good looking. It’s nice. It’s better than a lot of other sporty cars, but it’s not jaw dropping. It’s acceleration IS however.Report
I doubt it will be able to combine real auto sexiness with the “thrill” of not driving. Part of the sex in driving is the control, the action, the (pointless) flooring it. If a car has 300+ HP and sleek lines people will want to drive it, not toodle at the speed limit w/o doing anything.
Important caveat; yeah some people will definitely enjoy some…cough…sexy time excitement with a car they don’t have to drive, either with a partner or just themselves. I doubt the thrill of …you know…will be enough to convince people to buy an auto auto nor will Uber want people to be doing that. Uber is going to want people feel the seats are clean and unstained, not sticky, if someone orders an auto uber.Report
ewwwberReport
“Would you like a Lyft ( coy smirk) big boy?”Report
Part of the “sex” in muscle cars is the pointlessly loud engine, with deliberately excessive engine noise.
Executive cars are quiet things, sleek and sexy.Report
New option on cars: Speakers to really play up that engine roar.
Possibly as a default option on a Prius, to simulate engine noise at low speeds to warn pedestrians.Report
as a walker in a neighborhood with Priuses, that might actually be a decent feature.
As a walker in a neighborhood with blind people and Priuses, that is an actual lifesaver.Report
Bicyclists too. I’ve had cars in electric mode sneak up behind me at traffic lights. It’s not a problem when we’re moving at any speed because there’s enough tire noise for me to hear them.Report
Priuses should make the Jetsons ‘deedle’ sound.
https://youtu.be/LnT1VgeXOF0Report
Really? Up to the point of imposing no fault insurance? Because I don’t think we can have driverless cars without no fault insurance. Sorting out liability in collision situations is too complex otherwise.Report
Yes, including that if necessary. (Again, the whole post is speculative, but I think the authors of this site will themselves pen the necessary arguments if they see systems working in countries with more lax requirements and we’re the ones missing out.)Report
Trial lawyers vs Silicon Valley vs automobile insurers vs health insurers in an all-out death match in Sacramento!
While intellectually I want Silicon Valley to win this one (and even being a good Democrat), I’m kinda rooting for everyone to lose. The only way to make the lobbying any worse would be to add a couple of public agency unions.
(Who knows, maybe the sheriffs’ union will see an angle and get in the mix.)Report
I fully expect they’ll keep their eyes on the prize:
How’s that for a first draft?Report
The first company that markets a road ready driverless car will be first in line for the contract for Auto Pursuit Driverless Cop Car that will safely follow and overtake a rouge vehicle. That will be a heck of a program with all the robocop references implied. It would certainly be safer and give the cop time and focus to make radio calls, turn on ( or off) his body cam, etc, etc.Report
You, sir, are a master of your craft.Report
Dang Burt, that’s worth about 25 points of Sesame credit right there.Report
I’d expect the driver less cars to be full of cameras recording the car’s conduct: front, side and rear view. That’s give some clue as to who’s car was at fault.Report
Regular cars are starting to be like this. My mother-in-law got a new car a few years ago (and I can’t recall the make or model, sadly) that has so many cameras that when she goes to back up, there’s a screen that switches to a top down view complete with markings to indicate distance. (You can also turn it on with a button press)
It makes turning, parking (parallel or otherwise) very simple. It models the car perfectly, with live images of everything on all sides in a very simple format that makes it clear exactly where your car is in relationship to everything within about 15 feet.
Also has radar and a few other things in case you’re not bothering to look even at the magic screen when backing up, that gets increasingly loud to let you know you’re going to back into something, moron.Report
Yes, as the requirements for rear crash safety were increase, manuf designed smaller and smaller rear windows, reducing LOS. Then after a few kids got run over by suvs being back out of driveways by adults not paying attention, back up cameras were made mandatory.Report
*snort* s’not it at all. I actually know the person who lobbied for this one, and the arguments weren’t at all about dead children.Report
So what WERE the reasons? Also, are you taking exception to my explanation in any other way?Report
SUV’s are giant vehicles, generally in the hands of people whose experience was in smaller vehicles, and they’re generally quite high up off the ground.
Unless the entire back was glass, including the bumper, you can’t see anything under about three feet tall.
The cameras were added mostly so people could back the monster’s up, as they were unfamiliar, and so they could see bikes, toys, and small children that were hidden by the back of the car.
The average mini-van has far less of a view out the back than any SUV, and they’ve been around decades.
It wasn’t the window — it was the car height and the trunk panels, and owners who didn’t want to crush timmy’s bike or ram into another car as their unpracticed hands backed it out of a parking space.
But by all means, let’s blame regulation. That fits the narrative we like, right?Report
1) Cost/benefit analysis. Once the cameras got to be considerably less than a thousand dollars, it was a pretty easy sell.
2) When driving an unfamiliar car, a backup camera is a freaking lifesaver. Particularly if you get charged if you ding the damn car up. (the lobbyist drives Zipcars).
3) Yes, you do have the insurance and “It’s a bad idea to kill people” angle, but really, the argument was “This costs less than 5% of your car’s price” [and is the first safety feature in a while that actually concentrates on saving lives, rather than paintjobs — the more “safety” you put on cars, the harder they are to accelerate/steer out of trouble, often enough].Report
To be sure! That will create more information, of different and generally more reliable quality,than eyewitness recollections.
But there are also more players involved in the liability equation. We are going to be a lot more concerned with in-case-of-collision programming, hardware, data corruption, maintenance and mechanical failure, and other issues like these. In addition to vectoring out lines of impact, determining foreseeability and avoidability, and prevailing driving conditions as we do now.
There’s going to be more players in the stew of how a collision happened and less clarity as to how each player contributed to the chain of events leading to a collision. Not saying we can’t sort it out or that there won’t be information to sort. There will be more of it and it will be a more complex task to do. The end result will look more like a product liability suit than a negligence suit.
Product liability works on a strict liability (no fault) basis. So we will need an insurance regime that fits the applicable law.Report
Who actually wants this product, and why do they think it’s a good idea?Report
Me. I really suck at driving, and would like a car that would do it for meReport
I would embrace driverless cars with pleasure and joy bordering on extasy. The American car obsession has always baffled me.Report
I would welcome a truly autonomous car, but want nothing to do with a mostly-but-not-quite-entirely autonomous car where the human is expected to be constantly alert to the car’s surroundings and prepared to take over on a second’s notice. It would be hellish trying to maintain that level of concentration. I don’t believe for a moment that I or other drivers would actually manage it, and wackiness will inevitably follow.
I also am deeply skeptical of a truly autonomous car happening any time soon. A car that can drive itself on a freeway? Sure. I believe that. Well, mostly. I notice that the question of inclement weather induces mumbling and shuffling of feet and changing of topics. I’m not saying that the problem is insurmountable, but I sure don’t see people talking about how it will be surmounted. Instead I hear about cars driving around Mountain View.Report
Humans are already shit at paying attention to things. I don’t think this will make them worse, particularly if the computer can reduce reaction time. “here are three choices, pick the right one”
The question of inclement weather doesn’t induce mumbling and shuffling of feet. They’re testing in Pittsburgh, Pennsylvania, home of enough bad weather (and low light conditions) that you can routinely expect certain streets to be blocked off by police in the middle of winter. (Others, due to ice, have homeowners set up video cameras to catch the inevitable crashes).Report
It would have been more accurate to characterize it as mumbling and shuffling of feet every time I ask the question, and discreet silence on the subject in the press release puff pieces I see in the media. I would welcome a pointer to a discussion of the subject.Report
Well, as this is one of my meatspace friends, I’m afraid I can’t really throw you a link.
(and he’s harder than hens teeth to find online. Maybe try /b/, if that’s still functional?)
I think the way he’d put it (as any decent developer would) is this:
1) You get a decent set of working conditions, and create a system that works for them.
2) Then you work on the exceptions. Weather is just one of them, after all, and one that’s hard to control for (though that’s why they’re testing in Pittsburgh). You can parade trick-or-treaters around any day of the year, after all. Snow is a rather limited quantity.
If you’re concerned about visibility/etc in snow conditions, you could always ping the people selling the goosinator… (that’s not a wheeled vehicle, mind, and has significantly more freedom of movement).Report
I think this vastly overstates the ease with which the new players can get into heavy manufacturing and supply chains and understates the amount that traditional auto manufacturers have done in laying the foundation for greater automation. Things that are common in BMWs, Audis and Benzes like adaptive cruise control, lane change warnings (or automation), automatic braking, self-parking etc. are automation and are highly regarded by their users. The car companies also understand the market they serve and the regulatory environment.
Think about the Tesla fiasco with people sitting in the back seat while it was in Autopilot mode. Tesla tried to blame the drivers for this, but an entry level Ford Focus will beep if it detects a passenger in the front seat and the seat belt isn’t buckled. The old guard know how to take baby steps and not freak out the public or regulators. This doesn’t necessarily mean that they will win, but it means that they can’t be ignored as potential winners.Report
So here’s my example of a hard problem for autonomous cars. It happened to me at least once each session when I was working for the state legislature. Taking a day off during the session isn’t really an option, no matter what the weather forecast…
About noon it starts to snow. Hard. By the time I could leave at 4:30 or so, there’s eight inches already on everything. Still snowing heavily, so visibility is a couple hundred yards tops. All the main roads are slick as hell because the snow’s been packed down before the plows could get out. No lane or other road markings visible. Interstates are parking lots. Lesser-traveled streets are passable, if you know how to start and stop in that much snow. Successful navigation means knowing not just where the roads go in an x-y space, but also knowing where the steeper bits are. I got home in an hour (my neighbor spent four from roughly the same starting point), but that was possible only because I spent years and years learning how to drive in the snow.Report
You should see Brookline. At a guess, they’ll simply avoid the steeper roads, or take them very, very slowly (and I say this coming from Pittsburgh, where steep roads and bad weather mean black ice and people crashing. When they test in Pittsburgh, they’ll get this one). I mean, if you aren’t going to be driving, do you really care if it takes an extra half hour in bad conditions?Report
Add to this that the lane markings are irrelevant. You actually have de facto temporary lanes defined by the traffic that has already passed through. Trying to follow the marked lanes would be disastrous.
Here’s another for you. My daily commute over country roads includes one bad spot when it is icy. It is a dip down to a bridge over a small creek then up again, with a slight curve to add zest. You would barely notice any of this in clear weather, but with ice on the road it can become quite exciting. In really bad conditions, there will be a line of cars stopped before the dip. Once you get to the head of the line you stop and wait to see that the car ahead of you makes it back up to the other side. The idea is to work up as much speed as you can going down, without actually losing control, and then use that momentum to help the limited traction your tires have going up. If you don’t make it, then guys from the waiting cars will have to get out and push you: very embarrassing. Once I figured out the pattern I try to take other routes, but sometimes this is the least bad.
So when I hear about fully autonomous cars that are so wonderful that they don’t even have auxiliary controls, I wonder how they will deal with that spot. Or, when I hear about almost fully autonomous cars with auxiliary controls, I wonder about the people with virtually no experience driving, confronted with this situation.Report
A while back I wrote here with my questions about driverless cars. Something along the lines of this question was one of them. I don’t recall a satisfactory answer, but if I am misremembering I don’t doubt that someone will refresh my memory. (Oh, and five minutes? My daily commute includes a stretch of road like that, which many cyclists use, that runs about five miles. Those bicycles aren’t going 60 mph.)Report
wount these cars be hackable?Report
Probably, but only because it will be so by design.Report