YouTube Transcript:
Can You Fool A Self Driving Car?
Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.
Share:
Video Transcript
Available languages:
View:
I'm in my Tesla on autopilot going 40 mph towards a fake Wile E Coyote Roadrunner painted wall.
Please stop.
Please stop.
Holy...
- And I'm doing this to see if Tesla's autopilot can be tricked.
Because it famously only relies on simple cameras to navigate the world as opposed to much more
expensive tech.
In fact, we're out here today to run a bunch of tests in extreme conditions to see how those
simple cameras on a Tesla stack up against other fancy cars that use more advanced technology.
- Wow!
- And the story of how I ended up as a crash test dummy in my own car...
- Oh gosh!
(screams)
- started 6 months back at Disneyland as I was sneaking onto the famously pitch black ride called
Space Mountain.
And while I might look a little plus size in that jacket, as you can see from this infrared camera
shot, that's no ordinary jacket because it's hiding some of that fancy car technology.
- Yeah baby!
- 43 million people ride Space Mountain every year, but unlike a famous ride like Colossus, Space
Mountain is in near total darkness, meaning no one knows what the ride actually looks like.
Which, as far as I'm concerned, needed to be fixed with a special undercover operation.
- True story, I have been wanting to make this video for over 20 years, way before I had a YouTube
channel because I'll come to Disneyland and just be like, someday the technology will be there so
we could map out what Space Mountain actually looks like.
So this is a very big moment for me.
- It's common knowledge every good undercover operation worth its salt needs a 3-step plan.
For me, step one would just be getting into Disneyland with my tech through their famously
tight security.
And if I managed to do that, step 2 would be making it through the park, avoiding any security
guards on my way to Space Mountain.
And finally for step 3, I would need to sneak onto the ride itself while avoiding any suspicion to
make my childhood dream come true.
And if all this seems like overkill, let's just say I have a bit of history
- First row, please put your phone away!
- getting kicked out of amusement parks in the name of science.
- I follow you, man, why are you doing this to me?
- For step one, I took an oversized jacket and made a little pocket inside where I could hide my
secret tech inside it.
- It's like the slight limp shuffle walk that really sells it, I think.
- Because they make you walk through metal detectors, I disguised and temporarily placed my
device with a bunch of gear in my camera bag and just had to hope it wouldn't get discovered.
- Can I walk through?
- We'll just wait until your bag is fully screened in case you have items that are not supposed to be there.
- And even though this moment of truth was terrifying, I knew I just needed to play it cool.
Like that.
- Excuse me can you not record?
- And the security was in fact extremely tight.
But our alibi was tighter.
- All right, we're in 1 down, 2 to go.
- Once inside, I immediately made my way to a bathroom to properly suit up.
Now I just had to make it all the way over to Space Mountain without arousing the suspicion of
any of the park's security.
Which is a lot harder than you might think because it's a well kept secret Disney employs a bunch of
security officers dressed up in plain clothes as normal parkgoers.
So I could trust no one.
So in an attempt to blend in and look like a normal park goer, I took a corn dog break.
- We're doing this for the integrity of the mission.
And not because they're incredibly delicious.
- Back on track after a close call or two, I had made it inside the building.
- All right, 2 down now for the hard part.
- Now that we're inside, it's probably a good time to explain why my jacket looks like this when you
view it with an infrared camera.
Because that is a really tiny portable LiDAR scanner.
So it's shooting out 640,000 laser pulses per second to fully map the surroundings.
Now as for the legality of this, if you check the back of your ticket, it says you have to obey all
the park rules.
And if you check their website for the rules on forbidden items, it says you can't bring in shoes
with wheels or stun guns or cremated remains.
Ew.
But you'll notice the forbidden items list makes no mention of chest-mounted LiDAR scanners.
So technically, I'm not breaking any laws by precisely mapping out this entire ride down to the
nearest half inch.
- Don't blow this Rober!
We got this!
- After sneaking my way past the last of the ride security, I was finally in place.
And this is probably a good time to point out that the final boss here in this control room watching
all the infrared see-in-the-dark cameras was the one I was most afraid would shut down the whole operation.
Because when we tested this out to see how inconspicuous the LiDAR would be, as you can see here,
not very,
as it lights up the whole place like a Christmas Tree.
By the way, you should know LiDAR stands for light detection and ranging, whereas radar is radio
detection and ranging.
Basically, it's like having a bunch of laser pointers inside the shell shooting all around the room.
And each time you fire one of them, you then see how long it takes for you to see it hit the object
it's pointed at.
And since we know the speed of light, if we measure that time, it will tell us how far away
the object is.
For example, it will take twice twice as long to detect the laser point on an object 20 feet away,
as opposed to an object that's 10 feet away.
But how could the sensor possibly keep track of 640,000 laser pulses firing all out at once?
Well, it turns out, amazingly, they actually happen one after another, so the sensor only has
to keep track of one pulse at a time.
And this works because the speed of light is so fast in fact, if you slow time way down,
so the time it takes a laser to hit an object is four "Slo-mo" seconds, then you can wait 30
minutes on that "Slow-mo" time scale before sending the next pulse.
So it's actually really easy for the sensor to know exactly which laser pulse it's measuring,
even if that happens 640,000 times in a single second.
And with that, it was time for the execution phase of Secret Undercover Operation Mousetrap.
- Let's go!
(screaming)
- It felt so good to cross off a 30 year old bucket list item.
- (screaming)
- But just as I feared, as our exact cart pulled back into the station, they stopped the ride and
announced it was being locked down.
- All flights have been put in a holding pattern.
- Which meant I was about to have an unscheduled appointment with a bunch of dudes with muscles
much bigger than mine.
- Your flight will continue in...
- But it turns out it was an unrelated power issue which meant this was a mission accomplished.
- That's like a dream come true.
Before I even had a YouTube channel I wanted to do that.
- And the only critical oversight of the whole operation was forgetting they took those dumb key
chain photos at the end of the ride, leaving behind some very incriminating evidence at the scene.
- I forgot to put it away!
- That's so awesome!
- That's Harrison, by the way.
And he started the company that makes these super small portable LiDARs.
After wrapping out a few more rides like Haunted Mansion
- Let's go!
- and continuing my painstaking efforts to blend in, we headed home so Harrison could process all the data.
And a few days later he did not disappoint.
- This is from the LiDAR you were wearing on your chest.
- Get out of here!
- All of this is totally in the dark.
- Now, what's really cool is we could visualize what was recorded as the ride progresses.
And so as you can see here, you get a sense of what the LiDAR was seeing in real time as the ride
progressed on.
But what's even cooler though is we can now use that data to make an actual tabletop 3D model of
Space Mountain with the help of our army of 3D printers.
Which meant for the first time ever, this question mark's days were numbered.
- (extinguishes)
Got it!
- So while all the 3D printers were hard at work, we also reviewed the data we captured from the
haunted mansion ride, where we made two super interesting discoveries.
The first is that the initial room they take you in is really just a fancy elevator so they can get
you underground to transport you to the real ride, which actually isn't in this house at all.
Because when you go down this hallway, you're walking under this hill you see here.
So by the time you load onto the ride, you're totally outside the park.
But if that's true, there should be some evidence for this besides our scan, and sure enough, There
is, because if you look back at this shot of the hill, you can actually see just the corner of some
kind of large building.
And if you check that alongside Google Maps, you could see this entire massive, above-ground
warehouse where the ride actually takes place that matches up perfectly with our LiDAR scan.
Which is just super clever engineering by the super talented Disney imagineers.
And the second discovery is they use a lot of super thin, nearly invisible curtains they project
on that create a lot of the illusions like these flying ghosts.
And while your eye might get fooled, as you can see here, there's no hiding from the LiDAR scan.
- And so you can imagine there are some incredible real world applications for the LiDAR technology,
Like every day when your phone scans your face to unlock itself, or when archaeologists use it to
find lost ancient cities from the sky.
Or, for example,
(car braking)
... in cars.
- And as you might guess, this car automatically saved itself from crashing because it uses a LiDAR sensor,
which as you can imagine, would have saved Wile E Coyote a lot of pain over the years.
And the reason it works so well is because it's updating that point cloud of 640,000 laser
measurements every second to the brain of this car showing exactly what's in front of it.
In fact, it works so well, as you can see here, I can maneuver all around these obstacles in real
time with all the windows blacked out, looking just at that point cloud in a VR headset.
- OK, easy bike guy.
- My Tesla on autopilot, however, only relies on simple cameras and its image processing to
navigate the world.
So to see if that tech is just too simple we came up with a six-part LiDAR versus camera head to
head, face off, culminating in a history making first Tesla-versus-cartoon-physics test.
- OK, so we're starting out with the car with the LiDAR.
First test is this kid standing in the road.
Not sure why he's doing that.
It's very unsafe, so we want to be extra careful.
OK, hit it.
- The testing speed was 40 miles an hour.
- Should have put my seat belt on!
Which meant the LiDAR would have to detect the kid and then slam on the brakes at least 60 ft in
front of him, and it turned out...
(screams)
- That's all it needed.
- (laughing)
- Now it was Tesla's turn.
- This is a terrible feeling - driving straight at a kid, but this is for science.
All right, we are up to speed.
With just simple cameras the Tesla was speeding fast.
Oh no!
- But did detect the kid.
- Oh sh...
(laughing)
- Just not in time to fully hit the brakes.
- Oh no!
He was even a CrunchLabs fan.
- If you're a Tesla owner, there is a silver lining
because we were relying on the automatic emergency braking system to stop for the car,
and because it assumes the driver's paying full attention while fully driving the car, it only
needs the brakes when it's 100% sure there's a problem in order to avoid false positives.
So the alternative is to use autopilot, and that assumes the driver isn't paying much attention and
while the downside is you get way more phantom braking and false positives
- (laughing)
That was ridiculous.
- The upside is you're less likely to be charged with vehicular manslaughter.
Because you can see here on autopilot it actually stopped in time.
So I decided to be nice and call the score 1 to 1.
And then I'd be even nicer by using the more conservative autopilot on the Tesla for all the
remaining tests.
Such as this one where we now simulate the kid dashing out from behind a parked car using some
clever engineering,
giving the cars less than a second to identify the child and stop themselves.
- Here we go, getting up to speed.
Oh, a ball came out.
I wonder why that happened.
Oh, kid!
(laughing)
Another life saved!
- Now it was the Tesla's turn on autopilot.
- 40 mph, there goes the ball.
- And impressively, it stopped with plenty of room to spare.
- Good job, Tesla!
- Which meant we were all tied up heading into round 3, the fog round.
- Optically, with my own eyes, I can no longer see that there's a kid through this fog.
The LiDAR has no issue.
This will be interesting.
OK, here we go, a wall of fog.
- So we plunged through the fog coming to a very sudden brake.
- (cheers)
- After which we saw the mannequin was not only still standing, but it was casting a really cool
long shadow, because the lasers don't pass through solid objects, just like how you can cast a long
shadow with the sun because light doesn't pass through solid objects.
- I don't have high hopes here.
But would the kid look just as cool after the Tesla test?
Oh, oh, oh! Fudge!
(laughing)
- I mean, so it does make everything look sort of cool, but that wasn't the only shock.
- I actually hit the brakes there.
- That was on autopilot.
- The cameras didn't even hit the brakes at all.
- The only thing still on is the pants.
That was a bad one.
- Now that LiDAR had taken another W
- It's time to make it rain!
- The next test would see if the cars could spot the kid under a torrential downpour made up of maybe
too much water.
- This is really interesting.
See, the Tesla can see the kid, but as soon as it starts raining, the kid is gone.
- And it was similar in the LiDAR car where you first got a clear image of the kid and the shadow,
and once we started the hose,
- oh, you see all the water going in.
LiDAR might struggle here.
OK, here we go.
- And as the wall of water started, LiDAR seemed to not slow down at all
- Oh boy!
(giggling)
- until the last possible second.
- Another W.
Now I just need the rain to stop.
You know any good rain dances?
- (rain)
I could just back up.
- Oh, that's actually a good idea.
(laughing)
- Now the LiDAR surprised me.
It was time to see if Tesla could as well.
- Oh boy.
Oh boy.
(laughing)
- And sadly It did not.
And though I did everything I could,
- Yeah, he's no longer with us.
- So we moved on to the penultimate Round 5, where we had 6 of the brightest lights that money could
buy, simulate either sunset or sunrise, or a truck in the road having its bright lights on.
Would these lights keep the cars from detecting the kid?
- Blinding trucker light test.
Let's go.
- LiDAR could immediately see much more than my eyes could when the lights turned on.
- Oh, that's a bright light.
- The question remained if it would see the kid.
- Oh, come on, LiDAR.
(laughing)
- And it did it, no problem.
- It just always waits till the last minute.
It gets me every time.
That's very bright.
Time to be blinded by the light.
- And it's true, these lights were so bright, I wouldn't have seen this kid on my own.
And at this point, I really doubted that the Tesla could either.
But when it pulled to a stop,
- OK, we stopped for the kid with the bright lights, a W for the Tesla.
- And that W meant these were the scores headed into the painted brick wall ultimate grand finale.
Scientific papers have actually Been written on this exact scenario, debating theoretically what
the Tesla would do, but we were here once and for all to silence all the debates with cold hard data.
- All right, so LiDAR is perfect and Tesla is 3 for 5, but as far as I'm concerned at this point, this
one's the only one that matters.
- Oh my gosh, I can't believe I'm gonna do this.
All right, hit it.
- Now as you can see as a human driver, while that looks sort of convincing, the image processing in
our brains is advanced enough that we pick up on minor visual inconsistencies and we wouldn't hit it.
- And as for the LiDAR, it isn't looking at what image is printed on the wall.
- So this sort of just looks like a wall, which would make this the easiest test of the day, and
as I suspected,
- Handled it with no problem.
- So the question was, would the Tesla detect and stop for this Wile E Coyote style painted wall
that was hiding yet another child staring at nothing during what might be his last minutes on Earth.
- Only one way to test that hypothesis.
I was actually supposed to go but I chickened out.
(grumbling)
The car may not know it's a wall, but I know it's a wall.
Everything in your body says, don't drive into a wall.
Oh boy!
All right, here we go.
- And so I steeled myself and accelerated the Tesla up to the 40 mph.
And as the wall crept closer and closer without moving an inch,
- Holy crap!
- The question was if the Tesla would detect it in time to step on the brakes, and it turned out...
- Holy...
(screaming)
- So I can definitively say for the first time in the history of the world, Tesla's optical camera
system would absolutely smash through a fake wall without even a slight tap on the brakes.
- (laughing)
My heart is gonna beat out of my chest!
- (lauging)
Turns out my Tesla is more Wile E Coyote and less Roadrunner.
Sorry, little fella.
Looks like we lost two arms and a head.
Not sure that's salvageable.
I think I might Uber home.
- But that's not all folks, because while LiDAR proved itself the superior car technology, could
it actually deliver on the dream of 14-year-old Mark Rober and map out Space Mountain well enough
for me to make a scaled 3D printed model?
And after 30 years of waiting, I'm happy to report for the first time ever, this is what the Space
Mountain track actually looks like.
- Whoa.
This is so cool!
Oh, there's me.
There's me; little backwards hat.
Yeah!
I recognize every single one of these turns.
You go up right there and then you come down.
Woo!
And then it's all the right turns, right turns, right turns.
(screams)
Right here is where they take your picture.
(screams)
This is wild.
14 year old me would be so proud right now.
- This is it.
Space Mountain has been revealed.
- And thanks to Harrison's mapping technology, this means you can now actually see what it would look
like to write Space Mountain with all the lights on in CrunchLabs,
even if at this size you'd be way below the ride's 5 ft minimum height requirement.
And all of this is why I love being an engineer.
- You get to shape the future, potentially save a bunch of lives and accomplish even the most
ridiculous 30 year life goals.
If you're a teenager or an adult and you've always wanted to make and build cool stuff but just
haven't figured out that first step, this is it.
It's called the CrunchLabs Hack Pack, and it's basically a series of really fun programmable
robots that get delivered right to your door where we build it together and learn step by step the
kinds of engineering skills that go into making the builds on my YouTube channel.
And they all work with no programming required, but since my goal is to take you from wherever
you're at and level you up, you can easily hack the microcontroller brains of any of these robots
in a bunch of ways to completely level up the functionality.
There's also a community where you can share your builds or post questions, as well as an AI chatbot
named Mark Robot that will check your code for you and help you implement your most creative ideas.
On top of all that, each box has a chance to contain a platinum diploma.
If your box has it, congratulations, because college is now free for you or a loved one you
want to transfer it to, plus you get to come out here and brainstorm one of your own ideas with me
and my team for a day.
So if you want to enhance or even just take the first step of unlocking the really fun and
rewarding hobby of making stuff, just go to CrunchLabs.com or use the link in the video
description, where to say thank you, we're giving away that free box as an early subscriber special.
- Thanks for watching.
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
Works with YouTube, Coursera, Udemy and more educational platforms
Get Instant Transcripts: Just Edit the Domain in Your Address Bar!
YouTube
←
→
↻
https://www.youtube.com/watch?v=UF8uR6Z6KLc
YoutubeToText
←
→
↻
https://youtubetotext.net/watch?v=UF8uR6Z6KLc