Uber starts self-driving car pickups in Pittsburgh
techcrunch.com
techcrunch.com
The big rotating Velodyne thing has got to go. But so far, nobody who claims to have a suitable solid-state 3D LIDAR is actually shipping. Quanergy claimed a solid-state LIDAR suitable for automotive use [1][2] and showed it at CES, but never shipped. Their web site hasn't been updated since 2015. They were aiming at a price point around $100.
Advanced Scientific Concepts has a suitable LIDAR that costs far too much. Space-X uses an ASC unit on the Dragon capsule for docking. ASC announced they were getting into automotive in 2012 but never shipped.[3]
Somebody is going to do this soon. The technology works. It's a cost problem. Once the cost comes down, cars can have multiple LIDARs (at least fore and aft) near the top of the windshield and rear window. This gets rid of all that topside gear.
[1] http://www.quanergy.com/products/ [2] http://on-demand.gputechconf.com/gtc/2016/presentation/s6726... [3] http://www.advancedscientificconcepts.com/applications/autom...
November 17, 2015 09:00 AM Pacific Standard Time SUNNYVALE, Calif.--(BUSINESS WIRE)
* Quanergy Systems, Inc., will introduce in early 2016 the world’s first solid state LiDAR (Light Detection and Ranging) sensor for self-driving cars for less than $1,000 per car, it was announced today at the Los Angeles Auto Show’s Connected Car Expo.*
They're already six months late. The press releases keep coming out, and the hype continues, but where's the hardware?
American Scientific Concepts has working units which do what Quanergy claims to do and more, but they seem content to sell them to DoD and Space-X at a high price.
That sounds like shipping to me. You don't ramp up volume production as an automotive supplier until your customers are ready. Naming Flex and Sensata/TI as manufacturers is a fairly specific and strong indicator that they are ramping, unless you're suggesting securities fraud?
This sort of add-on thing is temporary. Another generation and cars will have an equally good sensor suite, but it will barely show.
And plenty of taxis in my neck of the woods have rooftop advertising, anyway.
On the other hand, it makes it clear the car is an Uber. Illegal parking and such will be a lot easier to ticket.
Objectively, tonnes of people are using it right now, have been for months (is it a year yet?) and there has been one incident.
In my opinion, it was developed by engineers probably a lot smarter than you or I, has enough sensoring for what it does, and despite not being flawless, might be better than no autopilot.
http://mobile.nytimes.com/2016/09/15/business/fatal-tesla-cr...
I'd also argue that there is a difference between incident and fatality. Incidents are also bad - even if they're not fatal.
1) They're basically duplicating the early Google cars with mostly the same sensor suite bolted on top and integrated in the side panels. This is good because it means that unlike Tesla, they're planning for fully autonomous.
2) Since the human driver is still responsible for switching lanes, it looks like this is very early stage testing and data collection.
3) I think it's great that they're taking the approach of giving customers a view of what the car is doing via the back seat tablet, and even encouraging them to share. This is a great way to get people comfortable with the tech.
4) It still saddens me that everyone is keeping their autonomous data proprietary. How much faster could development go if Google let anyone have access to their million+ miles of driving sensor data instead of needing to collect it all from scratch?
You basically suggesting Google to give up their(only?) leverage. Open source is all good, but if they can't make money out of these projects, you will see less of moon shots.
This seems like the kind of domain where the government can be useful. Negotiate terms to buy the aggregate data off everyone experimenting with self-driving vehicles and make it all publicly available for them all to use. Its a safety issue more than anything else, because we have borderline genocide going on every year with tens of thousands of roadside deaths, making cars safer is not just some side project for profits sake alone.
Why should the tax-payers subsidize private actor's R&D? I'm sure Google is amenable to partnership/sharing their data with anyone for the right amount or under the right terms; free hand of the market and all - no need for government "meddling".
Because this data is a matter of survival, where you can easily forecast at least several million vehicular fatalities in the next decade worldwide, it becomes relevant that the various states (I'm extending this globally since several European car companies are also engaged in automation development) it is paramount it be made accessible in the same way we seem to be finally having the epiphany that should have happened at least two decades ago that scientific research should be made openly available.
You make it sound like it is such a travesty or a great injustice, and I cannot for the life of me figure out why. Generating this data is not rocket science; it is fairly well-documented, the components are commodities and the process repeatable, one of the big challenges is that it is capital-intensive to scale. If the data is valuable as you postulate, raising the required capital should be easy, capitalism has been solving this problem for centuries.
> Because this data is a matter of survival, where you can easily forecast at least several million vehicular fatalities in the next decade worldwide,
No, it really isn't a matter of survival (at nation-state level you've elevated the discussion to). There are things that kill more people than vehicular accidents, so if your angle is "saving lives", then world governments really need to step-up their health game and meaningfully fight heart disease and cancer; hell, maybe they could even buy all the pharmaceutical research data and open that up to the public while there are at it (good luck getting a good price from GSK, Pfizer et al).
Google's push to expand their Cloud offering is also showing them trying to pursue non-ad based revenue.
-- disgruntled open source dev
I don't have the familiarity with what they're collecting (and how) to make any actual guesses on how you'd go about it.
Beyond that, no company is going to want to share their general data. There's too much time and expense behind its collection, and just sharing it wouldn't be enough; to make sense of it they'd have to share a lot of insights into their proprietary systems as well.
Then again, as someone who won't buy a self-driving car for as long as possible, it might be fun to learn the driving styles of different self-driving cars and use them to my advantage. E.g. If I know that tailgating a Prius which is going the speed limit in the left lane will cause it to move over in self-driving mode, or that a Mirai will always brake-check itself when it gets cut off, it might be possible to exert some control over them if it's advantageous in traffic to do so.
All the things you describe are as likely to cause accidents as wrong turns or over-acceleration. If Toyota can't get this right (my bet is they will) then they won't and shouldn't be doing autonomous cars, despite their "change of heart".
Pushing other peoples' cars around does sound kind of fun, but I hope we'll have something better to do than frustrate commuters.
How is that a bad thing?
1,300,000 people die each year in car wrecks and another 20,000,000 - 50,000,000 more are injured and/or disabled.
Google's motto is "Don't be evil" after all. Is valuing short term profits over millions of lives not "evil"?
>Don’t be evil. We believe strongly that in the long term, we will be better served — as shareholders and in all other ways — by a company that does good things for the world even if we forgo some short term gains.
-Larry Page
1) Google said don't be evil 10 years ago
2) I THINK that X is evil
3) Therefore, Google should not do X
Killing puppies is evil. Is not open sourcing sensor data they spent their money money to get evil? I don't think so.
If google opens the data (that I would argue is still incomplete), and irresponsible actors are given a means (via incomplete LIDAR maps, algorithms, etc) to mislead people already screaming TAKE MY MONEY, that would likely lead to a TERRIBLE image for self-driving cars if and when those race-to-the-bottom companies cause injuries and deaths.
I could foresee that those irresponsible actors could then scapegoat Google's data as the source of issues, whether or not that is true.
BUT I suppose injuries, deaths, and ensuing lawsuits are what prompts stricter licensing and oversight committees. I believe however that strict enforcement of self-driving cars will be an eventuality, no homebrew kits I'm afraid.
Running a private experiment to gather data.. that data is private. Sharing that data is opt-in, not opt-out. Google has no obligation to release it. (If they do, cool. But not evil of them not to).
Now it is valid to say the government should run more studies on self driving cars and share that data. But I am not sure any business would exist in this country if every private R&D effort they did got open-sourced.
> “Uber logs as many miles in 24 minutes as Google’s autonomous cars have logged in their existence,” they wrote. “While none of these miles are fully autonomous today, we just point out the scale of experience that can accelerate the development of the AI, mapping and learning for autonomous cars.” [1]
[1] https://www.bloomberg.com/news/articles/2016-09-12/google-ca...
Hmm… Let's try comparing 24 minutes of miles logged in Google Maps (and Waze) with Uber's data and see where they stand then…
>> “Uber logs as many miles in 24 minutes as Google’s autonomous cars have logged in their existence,”
Google's data is not limited to its autonomous driving project, despite what the cleverly-worded sentence implies. Off the top of my head: Google also has 3D LIDAR data from human-driven StreetView vehicles, GPS and accelerometer data from millions of mobile Waze and GMaps users.
https://www.tesla.com/en_GB/blog/master-plan-part-deux?redir...
As the technology matures, all Tesla vehicles will have the hardware necessary to be fully self-driving with fail-operational capability, meaning that any given system in the car could break and your car will still drive itself safely. It is important to emphasize that refinement and validation of the software will take much longer than putting in place the cameras, radar, sonar and computing hardware.
Even once the software is highly refined and far better than the average human driver, there will still be a significant time gap, varying widely by jurisdiction, before true self-driving is approved by regulators. We expect that worldwide regulatory approval will require something on the order of 6 billion miles (10 billion km). Current fleet learning is happening at just over 3 million miles (5 million km) per day.
Here Elon mentions a milestone of 6 billion miles. At their current rate, Tesla will hit that in 5 years, but clearly we can expect the rate to increase rapidly.
How many Tesla cars have with their self-driving feature? 20,000 cars? They'll have 100,000 cars RSN.
By 2020, we will have 2 orders of magnitude more data.
Sure, but so will Google. Their latest claim is > 1.5m autonomous miles and thanks to their Streetview project they have a ton more detailed mapping data than that.
Btw, I'm not saying that Google didn't do a great job, and make valuable progress. The game, however, is about to take a huge leap. Google was an early investor in Uber and many thought they were going to cooperate.
First, tesla collects zero lidar data. So no, they don't, because they don't collect the same kind of sensor data.
The second assumption is that the sensor data google is collecting, that is similar to tesla's, can only be collected by google's self driving fleet ,and not any of the other ways google has to collect data.
Because, as above, it's not the same kind of sensor data, this is also false.
Its business model was sacrificing someone's income in the middle to be cheaper than ordinary taxis. But with self driving cars, they can minimize the costs without hurting anyone.
Really really smart and huge move at the same time. Whoever pushed this idea in the company deserves a huge credit.
Is a total elimination better or worse than a reduction? I don't have an answer, and I see good arguments for each side...
This will happen sooner or later and for sure there will be companies and individuals that will be affected by the change, but that is not enough reason to stop the development. And this is not Uber specific topic either. If not Uber, some other company will one day work on this.
I hope Uber pay their taxes.
Think about this problem from Uber's perspective. More than two thirds of a fare go to the driver in most cases. Let's say eliminating the driver allows them to cut prices by 50% to account for purchasing a crapload of cars. Now imagine someone else builds self-driving technology (or licenses it) and copies the Uber business model.
A competitor building self-driving cars for-hire is instantly cheaper by a ridiculous margin. Uber would be overtaken by this competition in no time at all. This is unquestionably inevitable. Uber can't say "pay double (or more) so we can take the moral high ground and employ a bunch of fallible humans" because a.) humans are worse drivers than computers and b.) humans just cost more and nobody would pay to have a human driver.
Additionally, if Uber ever goes public, they're legally obligated to maximize return for investors. You can't keep human drivers around as a purely feel-good exercise.
A lot of business models do. Horse carriages eliminated income of (most of) manual porters, and cars eliminated income of (most of) horse drivers.
Unless you are a worshipper of Ned Ludd, that's how technologies work and will always work. Something new replaces something old, and people whose income depended on the old stuff have to look for different income.
I am pretty confident self driving cars are going to create more jobs than they destroy. Whole new businesses will be economically viable once the transportation costs are super low, traffic is less of a problem and when you don't need parking. On demand economy 2.0.
I'm definitely interested to see how this iteration plays out over the next few decades.
[1] http://www.ers.usda.gov/media/259572/eib3_1_.pdf [2] https://en.wikipedia.org/wiki/Technological_unemployment
In Pittsburgh you have a haphazard street grid with unconventional and often poorly maintained road conditions, hills with poor visibility, narrow streets, weird unpredictable driver behavior like making left turns in front of oncoming traffic as a light turns green (my friends call this the "pittsburgh left"), every sort of weather condition imaginable including heavy rain snow and ice, etc etc. A self driving car that can navigate Pittsburgh without ever needing driver input is truly autonomous
http://www.nytimes.com/2015/09/13/magazine/uber-would-like-t...
https://www.fastcompany.com/3046902/fast-feed/carnegie-mello...
This is a promotional video produced by Uber. It clearly shows the vehicle illegally entering a crosswalk containing pedestrians. That's not a great start, and not particularly good advertising.
(If the software was designed not to do this but failed to behave appropriately, then the situation deserves addressing.)
Also note Google's cars are explicitly programmed to exceed the speed limit in order to track with a lead car in order to get better info on road conditions ahead. [1] Technically illegal, but this both increases safety and is well within the bounds of normal human driving behavior.
To me it looks like they were fully inside the intersection before the car began the turn, so it was a violation (if very technical and unenforced).
>Sure there's a "letter of the law/intent of the law" distinction here, but I don't see this as a particularly egregious action, especially when compared to normal human behavior (human drivers do this often, and with much narrower margins).
Absolutely -- and as a pedestrian, I would prefer drivers punch through if they have sufficient clearance, rather than pointlessly sit there and watch me walk through.
But there's a difference between
a) "everyone has to do it even though it's not technically allowed", vs
b) "hey, let's showcase this as a prime example of our advanced technology in action"
I think it was wrong to do it as b), since it looks like some elderly people were cut off.
Otherwise, it would be practically impossible to make turns in a city...
I was always taught that from the moment the pedestrian steps off the curb, you cannot turn in front of them. They have the right-of-way.
In LA, if you did this you'd cause significant traffic issues.
If the intersection is wide enough that you can turn while maintaining a safe distance, then what purpose would such a law serve? Why increase congestion for little or no benefit?
But an officer, unidentifiable in the dark some hundred feet behind me, cited me for "not stopping at a red light" due to the fact my vehicle did not come to a complete stop.
Arizona has a state law, named after an unfortunate little girl killed by an asshole who blew through a red light, which makes any "red light violation" an automatic 3 points on your license and applies some certain minimum fine. (8 points triggers a suspension.)
Now despite the fact that there is an entire world of difference between (A) running a red light into cross traffic and (B) performing a California stop at an empty intersection, when I showed up to contest the ticket, the judge mocked me for my suggestion that, as I had endangered no one and not been reckless, I had not violated the spirit of the law and that the ticket should be dismissed. I had to pay the fine and accept the points. State law.
"Irrational" may well be the word I should have used, but it changes nothing. What purpose, indeed, was served by such a strict interpretation of the law?
-- Jean-Luc Picard
I agree, a law that serves no purpose shouldn't exist (the only purpose that one served was probably political). That's part of the reason why I find it hard to believe there are laws stating you must wait for pedestrians who are on the far side of the road. I searched but couldn't find them... and since such laws would do little or nothing to enhance safety while significantly increasing congestion (which is dangerous in itself), I think it's reasonable to conclude there are none.
People are probably interpreting the word 'yield' far too generously.
Interesting times, no doubt!
In my case, I like having a more aggressive driver as long as they keep things safe. Others prefer "10 and 2 while keeping it exactly at 55 mph".
There will be mistakes. People will die at the hands of self-driving cars. There will be backlash. There will be people who "hunt" autonomous vehicles (look at drones, and consider how much people value airspace vs driving etiquette). GLHF.
This. There are people today who "hunt" cyclists and Prii and roll coal at them:
Edit: I guess this is what trolling looks like when the troll hasn't discovered the wonders of the internet yet.
Ever been to a club where security outnumbers customers 5 to 1 or waiters huddle in a corner on a slow night in big groups of 10?
The complex senarios and infrastructure won't take decades. Google's driving car has performed insanely well so far for millions of miles safely.
I've seen them in most developing countries I've visited. Even on ~15 seat minibuses, where as well as selling tickets the conductor can help with luggage, and keep a lookout for potential passengers.
Bartender cancels last minutes? At least one of these 10 muscle head security guys likely knows how to mix drinks.
Not to mention that the difference between cost of AI equipment (fixed) compared to driver salary (order of magnitude lower) is not even close to developed nations.
That said, Turkey has a stupidly high traffic accident death rate.
So, yes, as the 19th biggest economy in the world, with 70+ million people, I think heavily subsidizing AI cars would definitely be in the country's best interest (and well within its means...Turkey already has at least one world-famous robotics laboratory). I'd wager economies number 20-50 are probably in similar situations. Yes, truly impoverished nations and regions of nations likely have more important things to worry about, but I'd still wager at least ⅕ the world's population falls into the "can't afford AI cars individually, but would benefit from gov't subsidization of such".
Another commenter mentioned countries where a significant portion of its citizens don't have basic sanitary conditions, and that's what I'm picture. And in the case of Turkey, I wouldn't call it a developing country using this definition.
> I'd still wager at least ⅕ the world's population falls into the "can't afford AI cars individually, but would benefit from gov't subsidization of such"
Agreed.
Yes, off topic, but I could've sworn that "developing" was the term used. Apparently, according to Wikipedia (https://en.wikipedia.org/wiki/Developing_country):
> In the 2016 edition of its World Development Indicators, the World Bank made a decision that there is no longer distinguished between “developed” countries and “developing” in the presentation of its data. Nobody has ever agreed on a definition for these terms in the first place.
So much for the vagaries of human language! ¯\_(ツ)_/¯
Going back to Turkey, I really enjoyed my time there and it didn't fit into the developing country stereotype since the middle class seemed to be doing alright and there were plenty of outdoor leisure activities (my personal marker of how "developed" a country is). If the political situation was better, I wouldn't mind living there at all.
It is just that if you live in rural India and are part of the 10% of the population that has no access to clean water Cholera is much more likely to kill you than an reckless driver.
Focusing on securing water in these regions is prioritizing safety. People are smart to focus on things more likely to kill them.
And going back to the original discussion, automobile accidents are an abstraction, a probable cause of death, versus something more definite like starvation.
I hope the Farmer's Almanac is wrong, or else we will get an interesting test of how self-driving cars perform in severe ice and snow conditions.
Examples: https://goo.gl/maps/gkJNuHFntgv https://goo.gl/maps/F5eTFCnDgV92
But in the summer and warm weather climates, this will come faster than anyone imagines.
Things tend to move fast when billions of dollars in profits are at stake.
I read in the spring their lone early prototype had 22 'coffee cup sized' cameras on the vehicle. By my count there are six 8 or 16 channel Lidar pucks positioned around the car not counting the big 64 channel spinning velodyne on top. There are at least 2 radar units, and likely also ultrasonic proximity sensors. Who knows what else, or what kind of compute they're packing in the trunk to manage all that data.
Sensor overkill!
Not that I'm paranoid personally, but these will give the owners a lot of information about peoples' movement patterns, a competitive advantage, an asset. For example, they could tell how many people wait at bus stops and sell that info to municipalities... They could look for trends in fashion on the sidewalks and sell it to clothing companies...
Even if they don't achieve autonomy, just putting cameras any vehicle and re-selling the footage data-miners could be profitable.
Even lane following is not an easy problem.
1/ interference. Having one lidar-beaming car going down the road works. What happens when there are many? I suppose it just means more illumination, but could there be an interference or "blinding" situation? What about pedestrian/pet eye safety? I sincerely hope that all are being considered now that this self-driving race is on.
2/ if Uber skirts around taxi regulations by being "just" an app that connects independent drivers with riders, how do they justify researching, producing, and deploying these cars with uber employees aboard?
I live in Mountain View and see self driving cars every day, often driving next to or past each other.
2. Long term, it's not clear to me that taxi laws would apply to cars that drive themselves. Short term, re this pilot project w Uber employees in the driver seat: taxi laws are local. Maybe they got permission from the city.
Let's say at time T=0, a few million Uber cars are parked or on the road. Users enter their travel plans into the app, and also indicate whether they are will to make an extra stop or more, whether they can drive a leg, or whether they desire a solo/express route. Then they either walk out to the parking lot and start driving, or wait to be picked up. So now, everyone in the Uber cars exists on this spectrum:
Express Passenger -> Driver for my leg -> Driver plus a little extra -> Driver plus a lot extra -> Full time Uber driver.
Let the market and scheduling algorithms work out getting people and cars to the right places are the right time. Now I don't need to own a car and we don't need to have highly functional self-driving vehicles.
These Uber milage credits are especially attractive to many casual drivers ( non Full time) and also good for UBER as they give only Credits(keeping 100% of fare cash) and it also put pressure on FULL time Drivers cut which is 80% today as more CASUAL driver pool is available in this new scheme .
Parking cars on People garage is economical compared on centralized location As it takes extra time to get cars back to neighborhoods where people need morning rides and also Uber saves on Real estate cost . UBER may have some centralized locations for Cars maintenance/repairs etc..
When 100% self-driving is ready by 2020/2021 ( both from Govt. regulation and technology side), All you have is Electric Self-driving Software Powered Cars (ESS Cars) parked all over in your neighborhood people Parking LOTS ( People get Mileage Credit for allowing their parking LOTS) and unlock and drive TO YOU from your mobile Phone .
[Ford says 100% self-driving in 5 years](http://money.cnn.com/2016/08/16/technology/ford-self-driving...)
some very rough figures: https://www.quora.com/How-much-will-a-ride-cost-in-a-Google-...
- Personal vehicles cost roughly 60¢/mile to own if you have a car payment, according to AAA -- or 2¢/minute, obviously keeping in mind that you pay for every minute you own the car, regardless of whether you use it.
- Taxis in NYC and Chicago run about $2.00/mi, or 50¢/min if you're going nowhere in traffic. - Uber depends on the city, but let's say $1.00/mi, or 20¢/min. Both Uber and traditional taxis have other fees they tack on, as a reminder.
So, the cost will be competitive initially: likely no higher than $1.00/mi or 30¢/min
with it getting as cheap as someone wants it to be for the sake of driving adoption. When you hit critical mass with an autonomous fleet, you could likely see fees below 20¢/minute (minute will likely be the dominant metric) to run a profitable model as we know it today.
With Mass adoption of Electric Autonomous vehicles, if it is offered at 40 cents/mile that that is cheaper than OWNING your own Car
---------------- 1/ Uber drivers in the U.S. collected an average $13.36 per trip, while Lyft drivers took $12.53 for each ride, according to a new report. http://time.com/money/3959091/uber-lyft-price-per-trip/
2/ Uber gets 20% cut of that ride price that is $3
3/ It seems average ride time is 15 minutes
[1] http://www.wsj.com/articles/inside-ubers-new-self-driving-ca...
http://www.npr.org/sections/alltechconsidered/2016/09/14/493...
I feel like a large amount of users use Uber to avoid driving when they've consumed Alcohol and should not be driving.
I wonder if this is just an interim stage.
2) GPS -- they know where the car is it all times
3) There is still a human in it. Once the human is removed, I suspect the car will only work if it has a continuous connection to HQ, and that will give it a constant GPS signal.
I do wonder, however, if a crash ends up happening, how will the blame-game go?
Let's hope it's for basic income...
> he
> he'll
> him
> two drivers in every car
I want to buy dictionaries for every reporter who writes about these issues.