An autonomous trip through Jerusalem in Intel's ride
Welcome to Jerusalem, home to some of the most aggressive and impatient drivers on the planet and home to Mobileye.
A company acquired by Intel last year for $15 billion.
The company's already made some 27 million cars safer with its adaptive safety systems and now all those cars are helping to make this It's the company's first fully autonomous car and we're going for a ride.
So here we will activate the system.
And so here we can see basically the camera that's looking forward which is kind of the main interface for the car.
We've got bounding boxes defining All the traffic, everything else around us.
What you see here is gathering from all the cameras around the car.
And making it a 3D view.
The car has 12 cameras in all, 3 that look forward at various levels of zoom, 1 that looks rearward, 2 on each side looking forward and backward diagonally, plus 4 more close range cameras, one on each side, that are used primarily for self-parking.
And that's it.
There's no radar and no light on.
Now we are in to 4 lanes now and merging into 2. now we must be assertive here other wise we will be stuck.
Going about 45 miles an hour now.
we know the speed limit based on traffic signs and also based on our
[INAUDIBLE] the [INAUDIBLE], so we know how human drivers are driving here.
And this gives us a very good idea of what is allowed speed, but what is the actual speed.
We're going through a tunnel, which means things are a little bit darker.
Does that make things harder for a camera based,
Yeah, and also note that there is no GPS here and we are still localized perfectly on the net.
Becuase we have visual landmarks that helps us to localize even without the GPS.
And are you reading the odometer from the car as well to help in this case, or is it just landmarks?
It's just landmarks now in the tunnel, there is no other
[INAUDIBLE] And now this route, even though it's pretty programmed, has the car been told anything about route?
Is there any special care for this route?
There's nothing special about this route in the driving policy, we potentially make it scalable by not doing any tweaks or tricks soley for this route.
So, we accelerated a little bit, you may wonder why.
The reason, and then we break.
This guy's not gonna let you in.
We will push our way.
That's pretty impressive.
So, the thing is that we have a right balance between being agressive and being safe.
So, we make this push, Like this, and then we sense, he didn't let us go, okay?
So we let him go, but the initial push signal to other driver that we want to go.
And then we will find the right moment to go.
And will the car cater its aggression to where you're driving?
Will it read the aggression of other drivers, too?
We make the driving policy parameterized.
And then, you can pick what driver you want for any part of the road.
Maybe you will choose a milder Drivers in California.
A more aggressive one in Boston.
Yeah, Massachusetts was exactly what I was gonna say.
So this can handle the Masshole driver, it can also handle maybe someone from New Hampshire or someone from Connecticut that's driving a little bit more relaxed.
So I notice we're not over taking this truck, we're going a little bit slower, why is that?
So when we approach the highway exit, we do not want to take an unnecessary risk and perform a lane change.
And then we will need to go back, maybe in dense traffic.
So we do assertive lane changes only when it is necessary.
Now we need again to make a change of lane in order to go back to Mobileye.
So we are waiting for the right moment, and now we accelerate.
The reason we accelerate there-
Quite aggressively [LAUGH].
Because otherwise the car behind us would not let us fit.
Okay, so again it's the communication without words.
Yeah, how much of processing and how much of logic involved is just to deal the uncertainty of humanity?
Part of the strength of our system is [UNKNOWN], because if you can explain many, many phenomenon based on few rules, then it generalizes to many crazy behaviors.
And we saw it yesterday.
When someone backed up, we simply stop, and then we perform a change of lane in 20 So basically providing boundaries between what is normal, what is aggressive, and what is unsafe effectively.
And allowing the car within that range.
When Shine says RSS he's not talking about really simple syndication.
He's instead talking about the company's responsibliity sensitivity safety definition.
Think of it as something like Issac Asimov's Laws of Robotics And instead of three simple rules, it's instead a 22 page PDF that attempts to define allowable behavior for autonomous cars.
When you talk about autonomous driving, you want to be safe.
In order to be safe, you want it to have redundant systems.
And if you fuse all the information of [UNKNOWN] and [UNKNOWN] in the low level, then you're not sure that you really have true redundancy When they show that the car drives itself completely based on cameras then completely based on radios and lighters.
Then you really get redundancy and you can be more sure about the safety of the system.
So when we get to a full production version of this car, it would have an additional radar [UNKNOWN] system, but the car would effectively be able to drive itself on one or the other.
So if one of them happpened to be shut down totally, you'd still be fine?
Wow, that's impressive.
Now here I will take control, you'll see that I will hit the brake, and then the system's off.
That was a little bit scary at times as rides in autonomous cars these days can be, but honestly,
That the car drives in an aggressive way was weirdly reassuring.
It's also really interesting that the car relies only on optical sensors, on cameras that cost just about 10 bucks each.
That makes the system much, much cheaper the systems from other companies.
And even when this does get laser and radar scanners added on, it'll still be far more affordable, which could give Mobileye and Intel a drastic edge.
In the autonomy race.