It never ceases to amaze me how many people have bought into the idea that Google is going to bring self-driving cars to the world. The leading car companies have been working on it for decades. And in some cases e.g. Volvo in Sweden are testing it at a serious, real-world scale:
They PayPal guy has a track record of delivering stuff.
Google, aside from a few select areas (Search, Mail, Android) has a big track record of conceptual BS and half-baked products that get pulled back or fail in the market (Chromebook, Motorola phones, Glass, Wave, Plus, etc).
Be fair. They're doing a lot of things, so obviously not everything will pan out, but the products that do are usually not just market successes, they become the market.
Self-driving is largely a big data problem, which Google is far better at than any car company. They also hired most of the big names from the DARPA Grand Challenge.
Google is focusing on a self driving car at NHTSA level 4 and plans to have such a vehicle available to the public between 2017-2020, while these other companies are focusing on level 3 and have no solid plans for level 4.
I thought it was about sensors and integration. Big-data? That would be open-loop driving (using GPS only to steer) which I desperately hope is NOT the approach they're taking.
I think all those sensors turn into big data. There's an enormous amount of information streaming off all those cameras, rangefinders, gps, accelerometers, and stock on-board car sensors (like traction control). Add hard realtime constraints for computing solutions, and you've got to handle a giant pile of unavoidable data to ensure safety.
Of course, I'm sure the data's culled a thousand ways to Sunday to make it manageable, but it's going to be like a self-contained industrial process, which can generate positively stupid amounts of data.
Google's LIDAR builds a 3D 1cm-resolution map of the world around it in realtime as it drives which is then compared to a 3D map of the route it received from Google when the trip started.
The downloaded route has been manually processed to include data about where signage is (precisely, in 3 dimensions, so the car knows where to find a stoplight without having to continually hunt), which routes it can take and at what speeds, exactly where the lanes are, rules of the road, etc.
The map also allows the car to know where it exists in the world to the accuracy of a centimeter, and would enable things like navigation of parking lots and garages.
I would consider having to build such a thing for the entire world to be a "big data" problem.
True; yet no human driver works like this. Humans respond to changes in the environment dynamically. Wind blow down the sign? Check for environmental clues - "Hey! I can see that's University Ave over there, this must be my exit". Traffic light out/obscured? Look at what the other cars are doing and recover. Pickup parked on the curb, partially blocking the right lane? Span lanes, and expect others in traffic to give you some leeway.
Open-loop driving is a terrifying idea. Not because it wont work 99% of the time; but because it will fail catastrophically 1% of the time which is way, way too often.
Which is Google's solution. But everything we've read today is that the big car companies are not using this approach. They are relying far more on the car alone to make decisions:
It's because car companies don't want to take the a big bet and try to do it all at once. That's why every deal Google has tried to make with auto-makers has fell through. Google wants to rush to the finish line and create a fully self-driving car. But the time-frame to be able to deliver that is fairly unknown due to a number of issues.
Auto-makers want to do smaller R&D projects that they can deliver to consumers within a few years of start (instead of a decade+). Things like adaptive cruise control or self-braking. These are things that were just a few years of R&D and are shipping now.
They are different paths to the same goal. We don't know the outcome yet, Google or the auto-makers may get there first. But we do know that auto-makers will have sellable features in the mean time while Google just has pretty marketing videos and demos.
What Google is bringing (that the article mentions) is the discussion about fully self-driving cars. This discussions means laws are being passed to facilitate the technology. It also means that the NTSB, insurance companies, and others involved in the auto industry are talking about the impacts of what a self-driving car means within our current system. So hopefully by the time they arrive, the rules and laws will be caught up with it. It also means consumers are talking about it and becoming aware of the possibility, which helps motivate others to work on the same projects. It may also mean that consumers are less wary of the technology when it actually does arrive.
It is all about sensors, algorithms, telemetry and maps. And pretty sure the car companies that participate in F1 would have a pretty good understanding of how to get all of those working together nicely. Not that it is a problem that only Google is qualified to solve.
Tesla sells in less than a handful of countries and given the amount of money they are losing with questionable success. Toyota, Nissan and GM have been selling electric cars en masse in almost all countries in which they operate i.e. most of them.
So yes. The existing car companies are best positioned to bring autonomous driving to fruition. They have already been doing it now with auto parking and accident avoidance. And let's not forget that have the supply chain, regulatory and marketing aspects sewn up.
Good! Tesla has pushed the big car companies to do what they could have done decades ago - build and distribute fully electric cars that are the same or better quality than non-electric cars. And, your questionable view that profitability for a company in it's growth phase defines success non-withstanding, they're pretty quickly eating their lunch at it too. They're going to continue to be forced to compete with Tesla because Tesla builds amazing cars that people want to buy. When they're able to sell them more places and build them at new price points (already happening), it's going to be fun to watch.
However, that doesn't mean the big car companies won't make cars that are just as good - the point is that now they HAVE to and that's good. Same with self-driving cars.
That's because self-driving cars aren't a car problem, they're a computer problem. Why would you think that a car company would be experts in machine learning? Doesn't matter how many decades they've been working on it, they just aren't domain experts.
Now it's possible they have since hired some experts, but a good 80%+ of all the well-known AI/Machine learning people I know of are working for Google.
Google have got the guys who literally wrote the book on AI.
Because the car companies ship tens of millions of cars today with auto parking and accident avoidance. They clearly know how to combine all the pieces in a robust and safe way.
And if 80% of the well known people in AI/machine learning are employed by Google then I assume that no interesting work is being done outside there. Pretty sure there are a few companies around who would disagree with you there.
>That's because self-driving cars aren't a car problem, they're a computer problem. Why would you think that a car company would be experts in machine learning?
Because they can hire expers in machine learning just fine. Just how they progressed from dumb cars in the 50's to modern cars full of real-time OSes, micro-computers and sensors and stuff.
Plus, I don't see Google as any kind of "experts in machine learning", with perhaps the exception of search.
Google having access to my email and web habits and showing me BS unrelated ads doesn't boost my confidence in their machine learning capabilities either.
Be fair; its both. Sensors and integration are part of the problem, and that's not Google's forte. In fact they had to hire the DARPA guys to get that expertise. Arguably car companies could do something similar to shore up their deficits in this field.
Furthermore, so many discussions conflate what are effectively advanced assistive driving technologies with truly autonomous vehicles (i.e. Robocar: Gp pick up little Johnny from soccer practice). The former is certainly interesting but it doesn't fundamentally change the nature of automobiles and driving.
To be fair, the leading car companies have been working on electric cars for decades, as well. And Tesla just comes in and takes the lead (same for space launches).
Company age has nothing to do with its ability to innovate (if anything, you'd want to squeeze the last penny out of your existing tech/products before offering something new).
I would add Project Ara to this list. The technologies being developed to make the modular phone (eg network on a board with UniPro) has the potential to drastically change the way companies design and fab all sorts of electronic devices.
Sorry but here is my "bold" prediction: Ara is the Google Barge of electronics packaging technology. There is no theory for how modularity will be a winner this time when it has always failed due to reasons that span technology, product management, and economics. Too many different problems to solve.
27 comments
[ 3.2 ms ] story [ 71.5 ms ] threadhttps://www.media.volvocars.com/global/en-gb/media/videos/13...
Google, aside from a few select areas (Search, Mail, Android) has a big track record of conceptual BS and half-baked products that get pulled back or fail in the market (Chromebook, Motorola phones, Glass, Wave, Plus, etc).
... ads, maps, cloud computing, cloud document storage & editing, analytics, fiber optics...
Be fair. They're doing a lot of things, so obviously not everything will pan out, but the products that do are usually not just market successes, they become the market.
Google is focusing on a self driving car at NHTSA level 4 and plans to have such a vehicle available to the public between 2017-2020, while these other companies are focusing on level 3 and have no solid plans for level 4.
Of course, I'm sure the data's culled a thousand ways to Sunday to make it manageable, but it's going to be like a self-contained industrial process, which can generate positively stupid amounts of data.
Open-loop driving is a terrifying idea. Not because it wont work 99% of the time; but because it will fail catastrophically 1% of the time which is way, way too often.
http://www.forbes.com/sites/haydnshaughnessy/2013/02/04/why-...
Auto-makers want to do smaller R&D projects that they can deliver to consumers within a few years of start (instead of a decade+). Things like adaptive cruise control or self-braking. These are things that were just a few years of R&D and are shipping now.
They are different paths to the same goal. We don't know the outcome yet, Google or the auto-makers may get there first. But we do know that auto-makers will have sellable features in the mean time while Google just has pretty marketing videos and demos.
What Google is bringing (that the article mentions) is the discussion about fully self-driving cars. This discussions means laws are being passed to facilitate the technology. It also means that the NTSB, insurance companies, and others involved in the auto industry are talking about the impacts of what a self-driving car means within our current system. So hopefully by the time they arrive, the rules and laws will be caught up with it. It also means consumers are talking about it and becoming aware of the possibility, which helps motivate others to work on the same projects. It may also mean that consumers are less wary of the technology when it actually does arrive.
It is all about sensors, algorithms, telemetry and maps. And pretty sure the car companies that participate in F1 would have a pretty good understanding of how to get all of those working together nicely. Not that it is a problem that only Google is qualified to solve.
Tesla sells in less than a handful of countries and given the amount of money they are losing with questionable success. Toyota, Nissan and GM have been selling electric cars en masse in almost all countries in which they operate i.e. most of them.
So yes. The existing car companies are best positioned to bring autonomous driving to fruition. They have already been doing it now with auto parking and accident avoidance. And let's not forget that have the supply chain, regulatory and marketing aspects sewn up.
However, that doesn't mean the big car companies won't make cars that are just as good - the point is that now they HAVE to and that's good. Same with self-driving cars.
Now it's possible they have since hired some experts, but a good 80%+ of all the well-known AI/Machine learning people I know of are working for Google.
Google have got the guys who literally wrote the book on AI.
And if 80% of the well known people in AI/machine learning are employed by Google then I assume that no interesting work is being done outside there. Pretty sure there are a few companies around who would disagree with you there.
Because they can hire expers in machine learning just fine. Just how they progressed from dumb cars in the 50's to modern cars full of real-time OSes, micro-computers and sensors and stuff.
Plus, I don't see Google as any kind of "experts in machine learning", with perhaps the exception of search.
Google having access to my email and web habits and showing me BS unrelated ads doesn't boost my confidence in their machine learning capabilities either.
Company age has nothing to do with its ability to innovate (if anything, you'd want to squeeze the last penny out of your existing tech/products before offering something new).