How Drive.ai Is Mastering Autonomous Driving with Deep Learning
spectrum.ieee.org
spectrum.ieee.org
One possibility is an automaker that wants self-driving car technology would agree to acquire the winner for a certain price, but this would already exclude most of the big players (Waymo, Cruise, Argo, Delphi, Uber).
Another possibility is the competition is funded by the big auto corporations in order to advertise who has the best technology, similar to racing events.
The biggest benefit from a public competition like this would be to improve public perception of self-driving cars so they're more receptive to the technology (and more likely to buy it).
They can't manage a computer-controlled transmission so far. (And we're not even mentioning SYNC).
This will likely be the inflection point that sees liability laws for computer software, because it's going to be a quite spectacular disaster.
Hope to see more competitions like this, and more participation from both smaller and bigger players.
I imagine if the little guys get enough press from things like this, the bigger ones will be forced to participate, or risk losing out on awareness of their tech's capabilities.
Not sure racing around a track is the best challenge or not but it's something!
Good luck. Let us know how it goes with a post here!
Do you have any long term hope that bigger companies will join self driving competitions like yours?
On the flip side, any poorly-funded upstart is unlikely to make achievements capable of swaying public perception -- especially in comparison to the autonomous car divisions of big multi-national auto companies.
TL;DR: We're probably beyond the stage of a DARPA challenge for autonomous cars. Billions of dollars of private money is already flowing.
Google, GM, Ford, Uber- they're at the point where throwing more money at the problem can't help it. Amongst start-ups that have yet to be swallowed up by a big carmaker or tech company, Drive.ai is by far the most exciting.
I mean, if a billion dollars (Cruise, Otto, Argo.ai) is the going rate for a demonstrably talented autonomous driving startup, the Drive.ai is already worth at least that much. They've got the most futuristic approach, and they're moving the fastest.
Right now all we have is marketing. This leads many to believe that Tesla is ahead in self driving tech, for example. You may or may not perceive that as a problem.
Regardless, I agree with GP that there is a space for independent testing. It could spur development further in the right direction. There is lots of investment now, though the target varies. Developing some tests could help consumers see what types of autonomy they want to buy, and give developers a chance to succeed in different areas, earning more unique awards than "best self driving car". We haven't established a public understanding of SDC testing criteria yet. I'm not clear whether developers agree on a set of tests either.
It's an odd state of affairs. Testing ought to be the first thing software engineers design.
California DMV does testing, but the results aren't sexy or widely known. Companies participate to varying degrees depending on how they feel it would affect their brand. Uber left CA and Tesla's participation was minimal. Also, the reports are only annual which seems slow these days.
We could use something like the league of women voters did for presidential debates, or like Netflix did for machine learning. Get the SDC community to scientifically show who really has the chops, and not just rely on marketing and branding!
I'm not sure how one would would even go about independent testing, now or ever. Gill Pratt, head of Toyota's SV research lab points out that to have statistically meaningful data about the safety of an autonomous vehicle, you'd need to do about 8.5 billion miles of real world testing on public roads. Of course, by the time one gets to the 8 billionth mile of testing, the early data will be irrelevant because whatever autonomous OS is being tested will have undergone considerable maturation and development in that time frame. I doubt there will ever be a huge amount of 3rd party auditing.
“You can apply for a permit to deploy when… you as a manufacturer believe the vehicles are ready to go,” says Bernard Soriano, Deputy Director of the California Department of Motor Vehicles (DMV). “The traditional system of not wanting to be sued crazy controls when manufacturers will do this.”[1]
[1] http://spectrum.ieee.org/cars-that-think/transportation/self...
That's a bit short sighted isn't it? Certainly the companies themselves have internal testing in order to help them judge, for example, whether some sensor data is useful or not.
I disagree strongly that the only valid test comes from consumer-driving. Plenty of things on the market go through trials before being sold to the public, be it toys, drugs or cars.
I'm not calling for regulated testing by the government. Just something cool that can get more people, even the companies themselves, excited about self driving vehicles.
George Hotz could do it. He knows the tech and has the popularity to get people's attention. Or, Chris Urmson, any others who previously participated in DARPA, or any existing SDC company.
Companies like McLaren and Ferrari made their brands by winning races and they make most of their money selling sports cars. Now imagine you're appealing to a demographic 100x larger and everybody wants self-driving technology, but they don't understand it well enough to know what to buy. There is a natural tendency for people to want "the best" and for people to associate prestige with trophies. If the event were properly marketed, I'd say the winner would make an enormous amount of money in sales. And it would definitely do wonders for brand loyalty.
That's why companies would be excited to participate. The public gets the advantage of a (hopefully) impartial competition to see how the technologies perform.
The first implementations of fully autonomous vehicles won't be sold to private consumers, they'll be fleet vehicles serving robotaxi networks, they'll be geofenced, and they'll require a fair bit of troubleshooting and maintenance.
Should we have a competition to see who offers the best robotaxi service? We will, and it will play out in the real world.
Investors would be interested. They're always trying to figure out which company has an advantage over competition. They want to know before it plays out in the real world, if possible.
They certainly have more incentive than other bodies.
It's been a long time since people have thought of insurance as cool.
“I think this is the first time autonomous driving has been approached so strongly from a deep learning perspective,” says Tandon. “This is in contrast to a traditional robotics approach,” continues Carol Reiley, cofounder and President. “A lot of companies are just using deep learning for this component or that component, while we view it more holistically,” says Reiley.
and
"This is why many companies working on vehicle autonomy are more comfortable with using traditional robotics approaches for decision making, and restrict deep learning to perception. They reason: If your system makes an incorrect decision, you’d want to be able to figure out exactly what happened, and then make sure that the mistake won’t be repeated."
“This is a big problem,” Tandon acknowledges.