Hacker News Reader

Comment by elmar | Hacker News Reader

A full self-driving car with no driver will take more than 30 years, and probably will only be possible on areas restricted to self-driving cars.

We will se first one to two seat fully self-flying vehicles than cars.

View on HN

On what are you basing this timeline? You realize that Google already has completely driverless cars roaming around public streets in silicon valley, right? There's even the story from an entire year ago about one of them getting pulled over by a police officer [1]. Every major auto manufacturer plus Google, Apple, Uber, and Lyft, and several other software companies are all working on this and they all have stated timelines of public-ready autonomous vehicles around a 2020 - 2021 timeframe. Will they handle every possible road condition in every possible location? No, but I'd be willing to bet money that in 2021 I'll be able to hail a completely driverless taxi/Uber in several major cities.

[1] http://www.cnn.com/2015/11/13/us/google-self-driving-car-pul...

I wouldn't put too much money on that bet. Adding twenty years might still be too optimistic.

Again, based on what? I'd happily put $100 on that bet today.

Based on the fact that no one is anywhere remotely near being able to do arbitrary pickups and dropoffs of passengers without a driver being present today. And that's totally ignoring any regulatory, legal, etc. realities which I expect will get worked out once the tech is rock solid--which it isn't remotely.

But they are near being able to do exactly that. Uber is testing and training its autonomous-capable cars in Pittsburgh right now, picking up and dropping off passengers on real routes every day. At their current rate of testing and expansion they'll have more than enough training data in the next two years or so. The biggest hurdle is regulatory, but there's so much money and so many lives (remember, human drivers kill people on roads every single day) at stake here that even regulatory approvals are moving quickly. See: Michigan [1].

[1] http://fortune.com/2016/12/09/michigan-self-driving-cars/

Small flying vehicles for daily travel seem insanely unlikely. The energy expenditure is wasteful, especially if you need vertical takeoff. Seems like not the best future for typical commuter traffic, I feel like.

on Vertical Takeoff you are completely right, personally I think we will see fixed wing electrical one to two seats self-flying planes.

The basic problem with VOTL is power/weight ratio of current batteries the maximum fly time you get is 15 minutes with no reserve.

If takeoff is still complicated, that would really make the typical commute take a long time anyway, no? You're still going to have to move between runways and a place of work somehow. Feels like it couldn't make sense for a majority of folk

for a one seat fixed wing aircraft you need a flat paved or grass surface the lenght of a court of tennis for takeoff and landing, no airport runway needed.

For 100% autonomy, you are probably right. But there are some useful steps along the way. For example, the first ones will likely just manage the interstate and long stretches of road during a commute.

The driver can "take over" in some form for parking, though even that may operate in a way that is completely different from today's vehicles. I expect to see a lot of things like this over the next 10 years.

Can you explain how you reach the 30 year figure? There's a lot of evidence suggesting that they will be available in some places under certain conditions in 3 years or less. Google and Tesla seem to think so, and they're the current experts on this.

Rain, Snow, Fog, Traffic cops, Construction zones, Yielding to emergency vehicles, Malfunctioning traffic lights, Newly repaved roads without lane markers, Pot holes, Railroad crossings, Standing water

I think much of this can be circumvented with real time data from apps like Waze.

However if the car is not able to find an alternate route, it may require the driver to take over, so it wouldn't be fully autonomous.

In either case, the car has to be able to reliably identify conditions in which it is necessary for the person to take over, which is definitely far away from being solved at a level that would be required to meet consumer levels of safety.

Apps can probably help in determining that there's an area where a driver is going to need to take over. I can even imagine a central database that construction sites, etc. need to make an entry into.

However, there still needs to be a driver who can take over which means it can't function as an autonomous taxi.

Considering that Elon Musk says we'll have full self-diving cars (from an engineering standpoint) in 2 or 3 years, this estimate seems very pessimistic.

It is unsurprising futurists working on pushing their cars that allegedly self-drive would tell you that the technology is closer to practicality than it is.

I consider Musk's timeframe a pipe dream.

Isn't Musk notorious for blowing way past his self-imposed deadlines?

By a couple of years, but not many decades, yes.

It could be 3 years or 30 years says Chris Urmson, depending on what you're talking about.

Elon Musk's predictions are consistently unreliable.

2-3 years in OK conditions, 30 years to cater for all the odd edge cases humans take cars. Bad human drivers, Indian traffic, cyclists, animals, narrow, unlit or poor roads etc. The other question is will the tech be there before the licensing/legislation.

Somewhere in the middle there's probably a point where self driving is demonstrably safer than human, even before they're "ready" for everywhere.

For a deep learning machine with LIDAR , there's only small difference between animals, indian traffic, bad drivers, and narrow/unlit/poor roads.

It's just a question of more training(altough a recent NHTSA? regulation says the planning stage cannot use black box techniques like DNNs).

And once self-driving is a business, i suspect we'll find ways to get said training data for cheap enough and fast enough. For example Tesla/Uber drivers as trainers .

You're grotesquely underestimating what fleet learning will do for traffic issues.

The way I see it (as a software engineer myself), is that each wreck that the AP did not prevent gets added as a test case in the AP unit tests that all have to pass. Also, each time a human wrecks, it could/will send the data to Tesla to continue simulations and ensure AP would not wreck in the same conditions.

How many of the wrecks, even fatal ones, essentially a different driver making the same mistake over and over? Likely millions. Real fleet learning would simply solve that problem once, and prevent thousands of fatalities as a result. As more and more AP capable cars hit the road, this simply will exponentially increase. Thirty years is laughable seeing how much computing has improved in merely the past 20. Maybe 5 tops assuming the legal hurdles are actually surmountable.

That's such an inethical way to do it. You're suggesting that we just let it learn from wrecks in the real world, instead of trying to prevent those wrecks by other means? This, the Tesla strategy, feels despicable.

*unethical by the way. I figure the idea is to learn something as opposed to learning nothing like we do today.

No it isn't, in fact it is how Tesla AP2 works literally today. What do you think Tesla's new "shadow mode" is?

http://www.theverge.com/2016/10/19/13341194/tesla-autopilot-...

They also have every single human driver wreck captured with AP capable hardware to make fully autonomous vehicles safer. The way they are doing it right now is brilliant.

Edit: As Elon has previously said, not releasing something that saves lives is actually unethical. If a single life is saved by AP software that would have died if the vehicle was driven by human meatware, not releasing it is the unethical thing.

Any attempt to prevent wrecks we do now are based on learning from past real world experience. All fleet learning does is automate that process and apply it to self-driving systems so that the learning is systematic and will help everyone using said self learning system instead of having to wait for new advice and/or new safety mechanisms.

Reply on news.ycombinator.com