Does anyone else think the steering wheel movements are unrealistically aggressive? The car is barely turning, yet the wheel is doing 60+ degree turns. Has it been calibrated to be this way, or Prius steering ratio is that relaxed?
Yea that did seem weird. Also as far as collision avoidance, shouldn't the car come to a complete stop if a van pulls out in front of it instead of just trying to move around it? What would happen if the van kept going?
Well, van stopped and yandex was slow enough to stop if necessary. Human would drive exactly the same way, I think. You won't go far away if you would give a way to every van who's trying to push his nose before you.
Remember, this is Russia we're talking about - or haven't you watched any of those dashcam videos? If it drove what we in the West think of as "normal", it would probably cause more accidents, not less.
Forget about that. This is a video of a car barely managing to properly drive in an empty parking lot. Google had released a video of a car driving through streets back in 2009. 8 years later they just now feel ready enough to do public tests.
Just wanted to bring some perspective about how big the journey is from doing a simple drive on video to having something that works well enough to be publicly deployed. We all saw what happened when Uber tried to publicly test way too early.
Too bad most roads in Russia are a joke. This will never happen until Russian infrastructure improves; albeit this tech will stay in the huge provinces, Moscow & St. Petersburg.
1.1mil city, 6th in Russia (Samara), city center (I lived not far from the place) [1] I would argue that's not a "good road". However, GP is right that this is Moscow-only (maybe Saint-Petersburg, too) anyway.
I live in Russia, and none of the roads outside of Moscow Ring Road have any markings at all. Yes, even big cities. At best the markings are painted in spring and disappear within a few months. Getting autonomous cars to work in these conditions is a huge challenge.
There's a famous saying that Russia has two problems: fools and roads, and it's just as true as when it was first coined.
Well, that's not true for at least two cities I lived in. St.Petersburg and Cheboksary. Cheboksary always had pretty good roads, even outside the city.
But - yeah, this kind of vehicles won't be a thing outside of Moscow for a quite a while.
I've always been skeptical of self-driving cars since they seem to be designed for Southern California conditions: clear weather and smooth, well-surfaced, well-marked roads. If Yandex starts out facing less optimized conditions, their self-driving efforts might produce something much more robust, safer, and more useful...
That takes me back. That's about the level of performance we had with our DARPA Grand Challenge vehicle in 1985. Driving around barrels is a basic test; we did that too. Ours worked OK at 15MPH, but faster than that we'd out-drive the LIDAR range.
It's much easier now to get to that level - good multibeam LIDAR is available, GPS/INS integration is common and cheap, and powerful computers that will survive in the automotive environment are available. Progress in vision has been huge, there's good software for digesting point clouds, and 3D SLAM works.
Most of the hard problems today involve dealing with other road users.
If you scroll down to the PhD theses names, there are some pretty well-known names that defended their dissertations as a part of the NavLab development.
(The Second) Grand Challenge, of course, was 2005. If I remember right, vehicles didn't have to deal with other road users until 2007 (Urban Challenge).
Well, I'm a driverless car developer (albeit for a very special use case, and definitely not in Yandex), please show me a decent "GPS/INS integration" which would be cheaper than, say, $1500 and would be of any use on the road and will be readily documented. With many sensors, the likes of Bosch make many good parts, but if you're not a giant auto maker, you won't be getting any documentation. And AFAIK even if you are a giant auto maker, Bosch will make the entire system for you, and you won't probably get access to implementation details.
Also, there aren't really any good high-performance industrial (dust and vibration proof) computers on the market, at least not to be bought easily. Please don't suggest NVIDIA -- starting with only a decent GPU is not an option, you'll need quite a few GFLOPs of general computing power to solve all that navigation, control, and, importantly, self-diagnostics.
By high-performance I mean at least a full four-core desktop i7 or Xeon, not some mobile CPU.
> Most of the hard problems today involve dealing with other road users.
The problem is, you won't get to dealing with those road users before you reconstruct the previous 20 years worth of technology in-house. It's not like there is some ready-to-use open source Stanley-level software.
We had a $5000 Novatel GPS. It didn't have INS integration, and required a subscription to a sat-based correction service. For the 2004 Grand Challenge, CMU had a unit which required 4U of rack space and air conditioning. There's been progress.
Of course there's been progress, or we wouldn't be able to even start working on this project with such limited resources as we have now — that's like $5000 for total hardware (still 4u rack though, +INS, +cameras, no corrected GPS), the main problem being the inability to buy something like tBOX [0] with a more powerful CPU. Though we are counting on both our system getting optimized and industrial PCs catching up in a year or two.
What I meant to say, though things are a lot simpler now, they are not that simple to discount the Yet Another™ lidar-based self-driving car.
Well, I'm talking only about additional hardware cost per car. To work in this space you primarily need a partner from auto industry, and that partner may be interested in automated driving only if the additional hardware cost is minimal.
Yandex has some of the best programmers in the world. This was a surprise to me, but in the past 5 years, I've been spending few months a year in Belarus with various engineering team and have been learning a lot about this region.
Yandex has it's own school within the universities where they recruit the best minds early on and train them.
I've heard from programmers who now work at FB. That they could pass google's interview tests, but could not pass the Yandex test at all.
Add to that, Yandex Taxi is currently the second most popular on demand Taxi service in Russia. They might also be able to easily bypass many regulations we face here in US and get to a lot more driving data than google/uber/apple/lyft might have.
What does that mean? I think, the self driving car might finally give Yandex team something to do with all the talent they've been recruiting. So I would not dismiss their efforts at all. They could become a real force very quickly.
There is local service called RuТaxi that had 1 million rides daily and recently merged with least one other service. I suppose it's could be less popular in Moscow, but according to RBC it's larger than Yandex, Uber and Gett together. Source in Russian:
It's actually better then Ya.Taxi in Moscow, yet less well-known. They have had fixed-price rides long before yandex did (they have for a couple of months).
And then there are some people who use Uber, and them I really don't understand. AFAIU Uber has solved a problem nonexistent in many countries, and now tries to push its brand where there were better options long before Uber started in US.
BTW, yesterday I've seen an Uber Eats-branded byciclist, now that might be competitive.
> They have had fixed-price rides long before yandex did
As far as I remember RuTaxi had it even 4 years ago, but I not sure when they got mobile app since I only ever used website.
> AFAIU Uber has solved a problem nonexistent in many countries, and now tries to push its brand where there were better options long before Uber started in US.
As person who lived in both Russian and UK / London I understand why it's makes sense. Uber model works in countries that have one of the problems: bad public transportation (like in Russia) or very expensive taxi (like in London). Also they solve usual tourist problems that local services can't solve because you have to invest time to know what local service in $countryname is good or bad.
I have no idea about other countries, but service provided by same RuTaxi for instance was worse: old dirty cars, sometimes even drunk and just scary drivers. Thanks for competition all my rides become much more pleasant in last couple of years.
Also in Russia rides was more expensive: rides that cost $8 (~250RUB) back in 2012 now cost $3-4 (~150RUB). Recently I used Yandex taxi for ~5 minutes ride and paid $0.7 (30RUB).
The number 1 (is in process) is combination of two taxi aggregator services (RuTaxi + Fasten) banding together [0]
However, on self-driving car aspect, they would not have major technical debt compared to Yandex who has the best mapping data in Russia as well as the technical expertise.
Most of the Yandex developers are not that good or not loyal to a company. This classes in universities is a scam: instead of teaching people of real world things, they teach how to work with yandex tech stack.
Yandex lost almost all good developers and it is continue even further. Companies rent offices near yandex and find yandex developers in the restaurants around this buildings and convince them to leave. This is very easy to make a decision to leave yandex for them.
Once i came to talk to yandex and ask - why they don't make good experience for mobile apps in their search results? Why they are not indexing anything related to apps? Answer was simple - user then will use this apps instead of yandex.
And such toxic ideas are pretty common in the company.
Talking about maps - they pushed out of the market one of the most awesome maps application in the russia. It had crappy UI, but it always make correct estimates and was able to learn your behaviour (more than 5 years ago!). While their app doesn't work at all but it couldn't because they need real drivers to make sane navigation. Then they pushed it to a taxis with a very simple trick - for taxis they made routes LONGER than it should be. This got them big fleet of a taxis. (sorry no proofs, but they are on russian and you won't understand it anyway)
I hope they won't be that successful with their crappy product on the roads (they failed in almost everything in last years) and it will be much harder to compete for smaller teams as everyone will say "huh, but there are yandex guys already!".
This video is a too dumb for a year of work. Voyage are better already, but they are a small team unlike yandex. Geohotz was even faster.
I agree with you regarding their maps and products. I stopped using google maps here in these area because of all the inaccuracies, but at the same time the UI/UX of Yandex is horrible. I do not understand why they could not create better product with better UX.
It's so frustrating to use their product and it's such a shame how they manage their team based on what you're telling me. I've been told by current CS/infomratic students here that, Yandex school teaches their student a lot about heavy math CS concepts instead of a specific language.
Wow, calling Yandex classes scam is bold accusation. Are you saying that the tech stack the Yandex is using is not from real world?
> Answer was simple - user then will use this apps instead of yandex.
Yandex is a business, not charity. It is the same reason why google will never use encryption in their email. Because they will not be able to scan for email content to display relevant advertisement.
Overall i find your reply very negative and zero value. You didn't pass yandex test, right? :)
> Overall i find your reply very negative and zero value. You didn't pass yandex test, right? :)
That cuts both ways - you don't see a problem with the needy practices of pretty much all major and many minor tech players, right? You don't get sick with all this mediocrity adorned with euphemisms, all this crap called awesome and all the greatness in life belittled? Do you honestly know not a single person that does quality work they can be honest about for a fair price that you would assume it's either "business" or "charity"?
> Overall i find your reply very negative and zero value.
The "argument" that "it's a business so hey" could justify anything, including robbing people, as long as you don't get caught. You know that providing value for society is one of the reasons societies let corporations operate at all, right? Modern business seems to think now that the bath tub is filled nicely (for them), there's no need for the plug anymore.
Machine learning lectures with Konstantin Vorontsov are one of the most brilliant course I ever seen. Konstantin is super smart, every topic explained in in-depth and with proper mathematical background. Video lectures of Yandex Data School - real thing, world class CS theory
Why add apps content into web search results? Using a website is much more comfortable than installing an app, checking its permissions, registering an account etc.
As a former employee of Yandex I totally disagree with what you said about 'not that good' developers and their loyalty.
It seems to me that the most common reason for leaving the company is immigration, which has nothing to do with Yandex itself. All my ex colleagues who left Yandex work abroad at Google, Facebook, MS and etc. Also I've seen many very good developers who cared a lot about their products. So my impression is somewhat opposite to what you described.
The OP doesn't seem to work there and seems like an outsider who thinks he knows how to run a company better than the people on the inside. This obviously misses a lot of moving parts and internal incentives (which yes may ultimately created a bad product). But I'd personally rather hear the experiences of people directly involved with the company.
I don't see anything wrong with training people on their tech stack either. That's not a scam, that's a highly valuable skillset if it gets you a job and experience. You can always learn other languages on your own, which is 10x easier after you know one.
Ugh, so much hate... It is okay to criticize the company and its products (nobody is perfect). But calling yandex data analysis school (I assume that is what you mean by "classes in universities") a "scam" is just disingenuous.
It is not a scam, it is basically a charity. First, it is free, no strings attached. You are even under no obligation to work for them afterwards (not even a moral one). The only price the students pay is that yandex becomes kind of an obvious place for internship, so many of the students apply (only then will they learn how to work with the "yandex tech stack", but that's the point of internship, no?) Allowing students to work with the yandex tech stack would be a hassle anyway - students would need to get access to the intranet, sign the nda etc... So these evil scammers teach them fundamental CS and machine learning concepts instead.
I guess the rest of the comment is similarly disingenuous and inaccurate.
Yandex taxi is really, really the best ride hailing service I have ever used. Wish them luck with self-driving cars, that would be a real challenge to start such a thing in Russia where drivers could be insanely careless about safety rules. For example a couple of times Russian taxi drivers asked me about WTF I buckle up for - is it a traumatic experience or something? :D
54 comments
[ 5.0 ms ] story [ 88.4 ms ] threadI hope this is an early prototype.
/s
Just wanted to bring some perspective about how big the journey is from doing a simple drive on video to having something that works well enough to be publicly deployed. We all saw what happened when Uber tried to publicly test way too early.
[1] https://a.d-cd.net/9661606s-960.jpg
There's a famous saying that Russia has two problems: fools and roads, and it's just as true as when it was first coined.
But - yeah, this kind of vehicles won't be a thing outside of Moscow for a quite a while.
It's much easier now to get to that level - good multibeam LIDAR is available, GPS/INS integration is common and cheap, and powerful computers that will survive in the automotive environment are available. Progress in vision has been huge, there's good software for digesting point clouds, and 3D SLAM works.
Most of the hard problems today involve dealing with other road users.
I think that date is off by two decades ;)
If you scroll down to the PhD theses names, there are some pretty well-known names that defended their dissertations as a part of the NavLab development.
(The Second) Grand Challenge, of course, was 2005. If I remember right, vehicles didn't have to deal with other road users until 2007 (Urban Challenge).
https://en.wikipedia.org/wiki/History_of_autonomous_cars
Also, there aren't really any good high-performance industrial (dust and vibration proof) computers on the market, at least not to be bought easily. Please don't suggest NVIDIA -- starting with only a decent GPU is not an option, you'll need quite a few GFLOPs of general computing power to solve all that navigation, control, and, importantly, self-diagnostics.
By high-performance I mean at least a full four-core desktop i7 or Xeon, not some mobile CPU.
> Most of the hard problems today involve dealing with other road users.
The problem is, you won't get to dealing with those road users before you reconstruct the previous 20 years worth of technology in-house. It's not like there is some ready-to-use open source Stanley-level software.
What I meant to say, though things are a lot simpler now, they are not that simple to discount the Yet Another™ lidar-based self-driving car.
[0] http://www.axiomtek.com/Default.aspx?MenuId=Products&Functio...
Yandex has it's own school within the universities where they recruit the best minds early on and train them. I've heard from programmers who now work at FB. That they could pass google's interview tests, but could not pass the Yandex test at all.
Add to that, Yandex Taxi is currently the second most popular on demand Taxi service in Russia. They might also be able to easily bypass many regulations we face here in US and get to a lot more driving data than google/uber/apple/lyft might have.
What does that mean? I think, the self driving car might finally give Yandex team something to do with all the talent they've been recruiting. So I would not dismiss their efforts at all. They could become a real force very quickly.
https://en.wikipedia.org/wiki/Gett
http://www.rbc.ru/technology_and_media/03/05/2017/5908a5de9a...
Paywalled English version:
http://www.unquote.com/cee/official-record/3004782/ufg-backs...
And then there are some people who use Uber, and them I really don't understand. AFAIU Uber has solved a problem nonexistent in many countries, and now tries to push its brand where there were better options long before Uber started in US.
BTW, yesterday I've seen an Uber Eats-branded byciclist, now that might be competitive.
As far as I remember RuTaxi had it even 4 years ago, but I not sure when they got mobile app since I only ever used website.
> AFAIU Uber has solved a problem nonexistent in many countries, and now tries to push its brand where there were better options long before Uber started in US.
As person who lived in both Russian and UK / London I understand why it's makes sense. Uber model works in countries that have one of the problems: bad public transportation (like in Russia) or very expensive taxi (like in London). Also they solve usual tourist problems that local services can't solve because you have to invest time to know what local service in $countryname is good or bad.
I have no idea about other countries, but service provided by same RuTaxi for instance was worse: old dirty cars, sometimes even drunk and just scary drivers. Thanks for competition all my rides become much more pleasant in last couple of years.
Also in Russia rides was more expensive: rides that cost $8 (~250RUB) back in 2012 now cost $3-4 (~150RUB). Recently I used Yandex taxi for ~5 minutes ride and paid $0.7 (30RUB).
However, on self-driving car aspect, they would not have major technical debt compared to Yandex who has the best mapping data in Russia as well as the technical expertise.
[0] http://freenews-en.tk/2017/05/03/on-the-russian-market-of-ta...
Yandex lost almost all good developers and it is continue even further. Companies rent offices near yandex and find yandex developers in the restaurants around this buildings and convince them to leave. This is very easy to make a decision to leave yandex for them.
Once i came to talk to yandex and ask - why they don't make good experience for mobile apps in their search results? Why they are not indexing anything related to apps? Answer was simple - user then will use this apps instead of yandex.
And such toxic ideas are pretty common in the company.
Talking about maps - they pushed out of the market one of the most awesome maps application in the russia. It had crappy UI, but it always make correct estimates and was able to learn your behaviour (more than 5 years ago!). While their app doesn't work at all but it couldn't because they need real drivers to make sane navigation. Then they pushed it to a taxis with a very simple trick - for taxis they made routes LONGER than it should be. This got them big fleet of a taxis. (sorry no proofs, but they are on russian and you won't understand it anyway)
I hope they won't be that successful with their crappy product on the roads (they failed in almost everything in last years) and it will be much harder to compete for smaller teams as everyone will say "huh, but there are yandex guys already!".
This video is a too dumb for a year of work. Voyage are better already, but they are a small team unlike yandex. Geohotz was even faster.
Wow, calling Yandex classes scam is bold accusation. Are you saying that the tech stack the Yandex is using is not from real world?
> Answer was simple - user then will use this apps instead of yandex.
Yandex is a business, not charity. It is the same reason why google will never use encryption in their email. Because they will not be able to scan for email content to display relevant advertisement.
Overall i find your reply very negative and zero value. You didn't pass yandex test, right? :)
That cuts both ways - you don't see a problem with the needy practices of pretty much all major and many minor tech players, right? You don't get sick with all this mediocrity adorned with euphemisms, all this crap called awesome and all the greatness in life belittled? Do you honestly know not a single person that does quality work they can be honest about for a fair price that you would assume it's either "business" or "charity"?
> Overall i find your reply very negative and zero value.
The "argument" that "it's a business so hey" could justify anything, including robbing people, as long as you don't get caught. You know that providing value for society is one of the reasons societies let corporations operate at all, right? Modern business seems to think now that the bath tub is filled nicely (for them), there's no need for the plug anymore.
Is that true? The sample curriculum [0] doesn't seem to support that.
[0] https://yandexdataschool.com/edu-process/program/data-analys...
It seems to me that the most common reason for leaving the company is immigration, which has nothing to do with Yandex itself. All my ex colleagues who left Yandex work abroad at Google, Facebook, MS and etc. Also I've seen many very good developers who cared a lot about their products. So my impression is somewhat opposite to what you described.
I don't see anything wrong with training people on their tech stack either. That's not a scam, that's a highly valuable skillset if it gets you a job and experience. You can always learn other languages on your own, which is 10x easier after you know one.
It is not a scam, it is basically a charity. First, it is free, no strings attached. You are even under no obligation to work for them afterwards (not even a moral one). The only price the students pay is that yandex becomes kind of an obvious place for internship, so many of the students apply (only then will they learn how to work with the "yandex tech stack", but that's the point of internship, no?) Allowing students to work with the yandex tech stack would be a hassle anyway - students would need to get access to the intranet, sign the nda etc... So these evil scammers teach them fundamental CS and machine learning concepts instead.
I guess the rest of the comment is similarly disingenuous and inaccurate.