Toyota Research Institute head says full autonomous driving is “not even close”

That being said, Toyota's (and Subaru's) approach is the smart way forward - add sensors and capabilities that augment the human, but leave the human always responsible and in control. In order for a crash to happen, the human AND the machine BOTH have to miss it.

Once the machine is good enough (and we figure out what "good enough" means), then it can take over driving. But not before.

Hello, Apple Newton, may you rest in peace. I'm not convinced we'll have handwriting recognition within my lifetime, and if we do I imagine that shortly after I'll go looking for Sarah Connor :)

And at least Siri is completely useless without a data connection.

I've felt the same way after watching various autonomous driving demos (like Tesla's[1]). But then I remember my experience of actually trying the same speech recognition software myself; under unrehearsed real-world conditions with edge cases and human mistakes, the technology performed terribly. Any "intelligence" I'd seen in the demo was essentially smoke and mirrors.

Granted, current autonomous vehicle technology incorporates a lot more artificial intelligence than those speech recognition demos ever did, but the challenge is also significantly greater (and more life-critical). Sure, your sensors and cameras might be able to read signage correctly in 99% of conditions, but when there's graffiti on the sign at night during heavy rain, all bets are off. The human driver may have difficulty also, but the human driver has real intelligence and life experience in a variety of domains, enabling them to make inferences based on more than just statistical probabilities.

In my opinion, prior to solving all the edge cases at an acceptable level using AI, we'll solve a different problem allowing us to sidestep many of those challenges; we'll incrementally start building (and converting to) smart roads where only smart vehicles are allowed. Obviously it won't be everywhere, but the most important routes will be covered, and your safety on those routes will be much higher than it would be on traditional roads with AI or human drivers.

[1] https://www.tesla.com/videos/autopilot-self-driving-hardware...

Toyota is way behind in the game. What do you expect them to say about the competition?

Computer speech recognition now has a greater accuracy than human speech recognition, but the types of use cases for talking to a computer are much less sensitive to error. Short snippets and phrases that are designed to get something done right now.

Compare ordering paper towels with Alexa, to asking a friend for a paper towel when your hands are messy. If your friend mishears you, they can just look over, see your dirty hands, and figure out you probably asked for a paper towel. Alexa has no such benefit, computers are held to a much higher bar.

> People thought intelligent voice-computer interfaces were just around the corner,

I would argue it isn't recognition accuracy that makes complex scenarios hard. I've seen context aware recognition being rolled out for keyboards recently (Android's keyboard does it now days, it'll correct the past word based on what the next word is), and it seems like Android does the same for speech reco, based at least on observed behavior during use, but I'm just guessing.

The real hard part is making computers smart enough to do useful things.

We are a long way off from

"Book tickets at The Altair for between 6pm and 7:30pm next Friday and add the reservation to my calendar and my wife's calendar"

being possible in all but a few contrived scenarios. (Now that said, the above scenario is getting easier and easier if you ignore voice, ML is good at figuring out emails that have schedules in them, and forwarding them to other people is now simple, everything is much better than 5 years ago!)

I don't think the technical aspects are as difficult for the 99.99% of the cases. It is the 0.01% of unknowns that will be difficult to overcome. But we can minimize that by modifying or building cities specifically for self-driving cars.

The most difficult part I think is the legal/regulatory issues.

This is why I see self-driving cars taking off in china/asia first before the US. They'll probably limit it to city limits initially. And when the technology is mature enough, broaden it to the entire country.

> What people don't realize is that full autonomous driving will require more than just faster/smaller/cheaper technical innovation - it will require the refinement of innovations that probably haven't even been invented/researched yet.

Like what? Object detection/recognition? Better sensors? Path/trajectory prediction? Vehicle control? Sensors and processing/unifying their data? The hardest part is letting all these building work together, and then produce a mass-production product from it, but we do have the base technologies, companies like Google, Tesla and Volvo have proven already that.

The problem comes down to software, and Toyota has the problem that it's a hardware-focused company. In the car-industry, just like in any industry, software will become more and more important, and Toyota - just like many other car-companies - simply doesn't have an answer here since they don't understand it.

https://www.luminartech.com/

They've been selling partial-electric cars since at least 2014. The Prius Prime allows for electric-only driving up to about 30 miles per charge. That's enough for my commute. I got 3500 miles on my first tank of gas.

They've hired other folks, too, who have real-world autonomous vehicle experience.

Google/Waymo are definitely leading the pack, but I don't think we'll see Level 5 in a production vehicle for at least five years, probably closer to 10.

I read the recent article about their test facility (https://www.theatlantic.com/technology/archive/2017/08/insid...), but 1. It's still California weather and 2. I've routinely seen worse intersections. What do you do in a rotary where you have to cut across multiple lanes of traffic to exit? If the autonomous vehicle waits for another car to give way, it will be there forever.

Also, as electricity demand and price goes up, and gas demand and price goes down, oil companies will quickly pivot to producing H2 by steam methane reforming with carbon capture and storage, to continue the returns on their billions of dollars of investments.

i really dont see autonomous driving as much of a race when the "race" is going to take 20-30 years to complete. cycles that long give everyone a chance to catch up and leapfrog each other

2022 is a good year if they come up with good tech. At the moment the market for EV's is small, tech is new and only Tesla really shows of something (they have to, because EV's are their only business). All big car makers are researching, from new diesels (double injector SCR's) to EV's & autonomus driving. And if they don't (or go too slow for some), then 1st tier suppliers like Bosch & Continental are doing it.

I can't speak to their autonomous technology, but Toyota has always been a leader in electric vehicles, from the early hybrids, to partnering with Tesla, to impressive long term improvements in fuel cells even after the U.S. Gov. Moved on to batteries. Arguably Toyota has more electric vehicle prowess than any other company. It is relatively small potatoes to change the source from open system cells to closed cells, especially when many of battery manufactures are also Japanese.

I thought both were already highly automated?

Cars will be automated because cars have to be automated.

Planes are mostly automated, except for liftoff and landing. They can probably automate those too ... but there's no point. You'll still need a pilot. Would you get into a plane flown entirely by computer with no pilot available to take over?

Trains are entirely automated. At least, the ones where there isn't a union enforcing some form of employment. I've been on a number of trains in airports that don't have any kind of operator.

Planes also have a very high safety bar to begin with, which cars do not.

Same thing about trains, there's very little value in doing so, and resistance in the form of unions in some locations. Some trains are fully automated though, here's an article with some more info: https://motherboard.vice.com/en_us/article/wnj75z/why-dont-w...

Frankly having some dude from Toyota, who don't seem to be very invested in this space, tell me he doesn't think it's going to work isn't very convincing. Chris Urmson saying it might take up to 30 years is far more convincing: http://spectrum.ieee.org/cars-that-think/transportation/self...

But that's 30 years for the entire planet, not until it works at all. WayMo is already letting people ride their cars in Phoenix - presumably one of the places with good weather and easy roads, so I think metro-level deployment in the next 4 years seems about right.

[EDIT]: And on the topic of trains. The MTA in NYC can't even seem to handle basic technical projects, like upgrading signaling infrastructure, or just extending a rail line a few miles without spending billions of dollars, so there's probably a decent chance that some of these are not automated due to sheer incompetence.

To give an example, my friend has to memorize the required speeds for each different section and turn in every train he runs, and has to manually adjust the train's speed accordingly.

Literally all of that can and should be done automatically, but the industry hasn't caught up with what is technologically possible.

[1]: https://en.wikipedia.org/wiki/Copenhagen_Metro

Actual aircraft can be fully automated for landing at least. In the US Army, I flew the Shadow 200 TUAV and it had a thing called the T.A.L.S. (Tactical Automated Landing System)[1] which does exactly what you'd assume it does by the acronymn. It is a monopulse radar tracking system on the ground with a transponder on the plane. You fly the UAV (Unmanned Aerial Vehicle) into a small area at one end of the runway and click the "Land AV" button (A.V. == Aerial Vehicle) and the landing is 100% hands off. You have the option to cancel the landing if wind or the parameters are off down to 10ft, but below that it is land or crash trying!

For trains, the safest ones are automated, but Unions (in the US at least) prevent most of the autonomous technology from being deployed. Where I live, in Chicago, we had a train conductor fall asleep and the train went up the escalator and crashed into the turnstiles at terminal 2 of Ohare[2]. A computer would have prevented this from happening as a computer would neither have been speeding or sleeping. In that incident, the automated breaking systems had never been tested and in fact were configured / setup wrong, so they both failed. This has been rectified since and additional barriers have been constructed. That being said, there are lots of fully automated trains[3], you just might be unaware of them.

    [1] https://en.wikipedia.org/wiki/AAI_RQ-7_Shadow#Design

    [2] http://www.chicagotribune.com/g00/news/ct-ohare-blue-line-crash-ntsb-report-met-20150326-story.html?i10c.referrer=https%3A%2F%2Fwww.google.com%2F

    [3] https://en.wikipedia.org/wiki/List_of_automated_urban_metro_subway_systems#Grade_of_Automation_4_Systems

Also, if I recall correctly, he said in certain conditions it's against regulations for a pilot to be manually controlling the aircraft, such as landings in high winds.

First, they have few human-error crashes; they're both significantly safer than automobiles, so unless automation can do better-than-human disaster recovery (not likely yet) there's minimal safety advantage. You have to outperform well-trained humans instead of random people to save lives.

Second, they don't take much time per traveller to operate. Both often have 100+ passengers per operator, making it relatively cheap to employ skilled human labor. Cars average around 2:1, so there's a lot more time being wasted per passenger-hour.

Third, cars have highly flexible usage. Planes and trains are generally moving people or cargo to approximate destinations, and travel to places design to store them. Cars spend lots of time arriving in storage-free areas - what if your car dropped you off at work and drove to a garage outside the city? And they do lots of last-mile transit - what if your car could go pick up your groceries or dry-cleaning without you?

Broadly, I think there's way more safety and financial incentive for 100% driverless operation of cars than any other mode of travel.

So, if it costs a billion dollars to make a good AI driver, then if you put it in a million cars then it's costs 1000 dollars each and you could eliminate lots of jobs where the main cost is the human (delivery, taxi) so providing a ready sales avenue for your self-driving car (or tech). If you put it in trains then you can maybe drive thousands of trains, and the human is a relatively low cost element of a train service so there's less room to expand.

I don't know if this is true, or even if the logic I outlined above pencils out with real data but it probably contributes in some way.

I've been on an automated city train as well, I believe in China.

Driver salary is 26% of the cost of trucking.

Driver salary is 52% of the cost of a taxi.

Those numbers are in the single digits for aircraft and trains.

We could replace pilots and train operators, technologically speaking. But the incentives are not (yet) large enough to overcome the regulatory and cultural barriers. We'd rather have that extra order of magnitude of safety, because it costs so little compared to other expenses.

The cost/benefit calculation for drivers is fundamentally different.

Just this week, I missed one of those variables and cost myself $500 in damage. This is something a backup camera and automated system can and already does easily help with, and is a major efficacy leap. Small steps like this can really add up.

What I would like to see next is a system that alerts the driver when the light in front of them turns green. I feel like I lose 5-10 minutes a week sitting behind people who are checking their phones unaware the light has changed.

We'll continue to benefit at each level: speed match, safe-distance maintenance, emergency brake, lane follow - there are already people alive who would be dead without them - you can find them in 5 seconds on youtube. The fact that articles have to keep trotting out same sad story of the bloke who drove into a truck while watching movies when he should have been driving tells you a lot about how safe these systems actually are - if there were dozens of counter examples, we wouldn't be hearing from the head of Toyota's RI, we'd be drowning in dashcam snuff films.

All that said, yes, I agree it's obvious that level 5 is a different beast to all of the 'easy stuff' that we have now: classic case of where the edge cases cover more area than the core of the domain does.

Please tell this to city governors who are so keen at throwing money to stupid start-ups peddling this nonsense at the expensive of other things, like proven public transport, for example.

Stuff like this being "patently obvious" to you should be an impetus to get the word out, because those for whom is isn't are the ones making dangerously imprudent policy decisions.

Again, not to invalidate the use of level 4 automation and such, but ya'll need to crush this fantasy of level 5 in 2 years, for all our sakes.

Yes and no. In any given Tesla article here you can also read many people claiming that "it's basically a solved problem and Tesla and Uber will be fully automated in just a couple of years", too.

Problems in AI can very easily turn into a black hole of money and data. I hope the large number of AI startups appreciate this.

AI is kind of a silly term, imo.

"I am driving at 45mph at [GPS Coordinates] [UTC Timestamp]".

"I'm braking [GPS Coordinates] [UTC Timestamp]".

Driving is quite a bit short of "same intelligence as humans"

Animals are very good at responding to specific situations with fixed action patterns.

Humans are very good at adapting to new situations by training new action patterns.

Self driving cars tread far more in the territory of machine learning than AI. AI involves ML, but ML doesn't necessarily involve proper AI. ML is essentially creating a framework of existing data that the computer can apply to similar situations. This is why you hear terminology like "training a dataset" in ML -- you are just telling it to act upon new situations in as similar of a capacity as possible to previous situations.

AI to me is a very different thing, in that it doesn't require structured inputs and outputs. Even as chaotic as autonomous driving is, it's still a structured system that takes inputs about road rules and surrounding objects and aligns the car with an outcome where it follows road rules and doesn't hit anything.

True AI is a machine that can reason for itself in unstructured situations, which would likely involve ML that is very good at making inferences about how existing data applies to tangentially related situations. I'm sure at some point there will be a distinction between AI that aims to create outcomes as similar as possible to human decision making (perhaps it could be trained by you to make decisions the way you would), and AI that aims to emulate human thought process literally down to the neurotransmitter level.

Levels 2-4 are the death zone (or "deadly valley"). These are the levels at which people will die unnecessarily.

One could make the convincing argument that Waymo's vehicles are already at Level 5; where they probably struggle (I have no examples or data on this) is probably in inclement conditions (rain, snow, fog, etc). That said, even in such conditions, they probably perform much better than a human driver.

For instance, most human drivers in such conditions - even when they struggle to see the road clearly - continue to drive anyhow, mostly blind, instead of doing the right thing and pulling off to the side of the road and waiting, which I bet is a behavior that Waymo's vehicle performs when it struggles beyond a certain level.

In other words, Waymo's vehicle is likely better at determining when NOT to drive, and acting on that determination, instead of being stubborn and irrational in the face of evidence to the contrary.

Driving cars is ultimately about learning and learning is asymptotic. A child typically progresses from not knowing how to control their vocal cords to 50% of their peak vocabulary within 5 years. It takes several times that amount of time to be able to capably work a white collar job (a feat many fail at), and decades more to be able to make a meaningful and attributable contribution to society (something almost all fail at).

The genome project is a terrible example because mapping progress is a second order effect to the actual technological progress of learning how to map the genome. It would be reasonable to project that once we know how to build a level 5 car that AVs will consume 100% of miles travelled within a very short period of time. But learning how to get to level 5 could very well be asymptotic in the same way that it takes decades of asymptotic refinement to turn an exponentially learning toddler into a PhD.

Waymo is actually testing for real in Arizona with selected people from the public.

Yes, they include an employee in the car to monitor or take over if necessary. But from the numbers I saw they reported to the DMV like a year ago, with Waymo's cars that happens surprisingly little. https://www.dmv.ca.gov/portal/wcm/connect/946b3502-c959-4e3b... Like very, very little. 635,868 miles and 124 reportable disengages. As of last year. Significantly improved now. In the rare case of a problem they can send a human to help.

If they had legal clearance or a waiver or something, they could TODAY just take some routes and times that had little traffic and not include the test/backup employee. Again, the number of times they report the driver needs to take over now is minuscule. They could then start charging for the rides. Then that would be a commercial deployment.

For them to make that widely available, at first for selected low traffic areas and times to reduce the risk of any incident or needing to send a human driver, is just a matter of scaling up their fleet and the legal issues.

Tesla is pushing very hard to the point of being unsafe to get their autopilot to actually work in as many circumstances as possible. They literally have been selling self-driving cars. The cars are collecting massive amounts of real-world data. They have hired many genius AI experts. Unless there are too many crashes or the company goes down in flames (which is possible considering how aggressively they are testing and releasing software even with issues), they will push to decrease the amount of driving the human has to do down to close to 0 as soon as is possible. Musk will try as hard as he can to get 75% to 100% there before the end of the year, because that is literally what he has promised. By the end of next year his actual conservative expectation is to be more than 80%.

As far as operating profit who knows, but there is no way that we will be waiting until 2022 for this to be deployed, at least in somewhat limited routes and times.

Autonomous vehicles have been virtually out of the "basic scientific research stage" ever since Darpa's Grand Challenge in 2005 and 2007.

While there is still a ton of research being done, most of it has been focused on refining existing solutions, or applying existing solutions into new problem spaces. For the most part, we know what is needed both in hardware and software to gain Level 5 capability, and for the most part we have working examples of all those functions.

What is left is refinement.

I also believe that Tesla is on the right path by using cameras for it's main sensor input (augmented by radar, and probably some LIDAR too). We already know that our road system can be navigated visually - it's inherently designed for this. In theory, radar and LIDAR shouldn't be needed at all, but just as they are useful augmentations for humans (in those areas that they are used - such as proximity detection on some vehicles), so they are also useful for self-driving systems.

That car had nothing but positive reviews from all corners of the internet. I am surprised Cadillac canceled its production and thought of making a coups instead of a convertible.

Not many are, but some are occasionally. I personally own one of the few existing Isuzu VehiCROSS vehicles. It started out as a concept car in the 1990s, then went into production virtually unchanged (only a tad under 6000 were ever made).

So manufacturers are already feeling the pinch, so to speak.

While (as you put it) "autonomous hire cars" will make the market for new car sales (and car ownership in general) smaller, I don't think it will completely go away, for a couple of reasons.

First off, such form of "public transportation" is still subject to "the drunk person who pukes in the seat" syndrome. You will quickly get vehicles on the road that are virtually rolling trash cans, and when you encounter one, you'll have to make the choice to either use it anyhow, or send it back and wait for another to come (potentially making you late to whatever it is you are doing).

Secondly, a service of "hire out your personal car" might become something; kinda like driving for uber, but not actually driving. Make money with your car when you aren't using it. Of course, you could also find your car trashed in the process (I'm sure AirBnB suffers from a similar problem).

Ford and other companies are just likely hedging their bets, plus whatever tech they do develop can also bring them a tidy sum by licensing it forward, or selling it off.

This is obvious even now with the Waymo v Uber lawsuit.

Consumers LOVE it. So even having some level of driving assistance will win you more customers.

Everybody's jumping in because the prize is ridiculously huge. This justifies jumping in even if you deem your odds < 1 in 100

It is much more comfortable to drive if the car itself keeps the same speed as the car in front of me and follows the road. I also feel a lot safer if my car would automatically brake if I'm distracted and a kid runs out in front of the car.

Note that being in the race just means they have a plan in place. Car makers partner all the time - it is common to see the same car from several different makes with only the logo changed, or an optional engine from someone else.

I expect that in the end there will be about 5 different self driving systems on the road that are shared between different cars. Designing a system is complex and expensive on the one hand; and not a differentiator that customers will pay for on the other hand: so it won't be worth the cost to continue to develop a system if you have it when you can buy from someone else. (though each might customize the UI as the UI is a feature they can sell)

Sounds about right. The same stuff market bubbles are made of!

I highly doubt it, though.

The thing is, while I believe they are already at this level, their technology is obviously not perfect. It never will be. Perfection in any space of technology is just not possible, because perfection is not possible in the physical world for anything. Perfection is a mathematical abstract at best.

Instead, what we can hope to achieve is "better than human drivers", and Waymo's vehicles have certainly achieved that, imho. They have logged more miles with their self-driving vehicle technology than most people will drive any single car of their own, with an accident rate far below that of the average human driver.

For the accidents they have had, most of them were when the car was in manual mode, and for the rest, they were low-speed incidents. None of them, that occurred while in self-driving mode (that I am aware of) caused any injury or death to any occupant of the vehicle.

I'll leave this comment with this video, to show what was capable of autonomous vehicles in what now seems like the distant past (and I honestly know it wasn't, but time sure flies with this tech):

https://www.youtube.com/watch?v=_piO849uRdI

Note that this was in 2010, using the tech in Stanford's Junior car (tech that would eventually lead to Google's and later Waymo's vehicles), and far from perfect; I can guarantee you that the systems used are much more advanced and better at control today.