Mark Rober just set up one of the most interesting self-driving tests of 2025, and he did it by imitating Looney Tunes. The former NASA engineer and current YouTube mad scientist recreated the classic gag where Wile E. Coyote paints a tunnel onto a wall to fool the Road Runner. Only this time, the t...
Mark Rober just set up one of the most interesting self-driving tests of 2025, and he did it by imitating Looney Tunes. The former NASA engineer and current YouTube mad scientist recreated the classic gag where Wile E. Coyote paints a tunnel onto a wall to fool the Road Runner.
Only this time, the test subject wasn’t a cartoon bird… it was a self-driving Tesla Model Y.
The result? A full-speed, 40 MPH impact straight into the wall. Watch the video and tell us what you think!
Rober is definitely a businessman out to make money and is very self-promoting and will accept just about anybody as a sponsor, but I can't think of anything he's done that's been out-and-out deceitful or political. And he really does have some engineering chops.
I think he's a good voice for this b3cause he's been so intentionally apolotical, and even my right-wing family likes his stuff.
Though my YouTube crazy engineer of choice is Stuff Made Here. He spends months between videos, but the stuff he makes is awesome, and he shows off a lot more of the actual creative process. And his fabrication tool collection is insane for a home shop.
Mark is a smart guy, I'm sure he walks great big circles around anything political, at least publicly.
His audience is everybody, aligning publicly with any kind of political flow is generally a bad idea if you want that to stay that way, because the only thing you'll likely achieve is shrinking your potential audience.
I would also be careful with the assumption that all conservatives agree with what's currently happening.
Observing a technical deficiency in a robotics platform requires political considerations. Even when a car drives into a fucking wall at 40MPH on camera, people are asking about the camera man's political party affiliation and not what's wrong with the car.
Unfortunately when the vehicle in question is created by a company owned by a man operating a government agency, it's a valid question. He could have just never made the video, but making one that directly opposes the narrative of people you'd expect the "camera man's" political affiliation to be seems unusual.
We still don't know for sure. That video will likely become one of, if not his top-grossing videos. The topic and timeliness are absolute fire.
I give him some credit, though. It's a dicey time to throw Musk under the self-driving bus while showing that alternatives don't have the same problem.
All these years, I always thought all self driving cars used LiDAR or something to see in 3D/through fog. How was this allowed on the roads for so long?
They originally the model S had front facing radar and ultrasonic sensors all round, the car combined the information to corroborate it's visual interpretation.
According to reports years ago the radar saved Tesla's from multiple pileups when it detected crashes multiple cars ahead (that the driver couldn't see).
Elmo in his infinite ego demanded both the radar and ultrasonics be removed, since he could drive with out that input so the car should be able to.. also it is cheaper.
I'd be very curious to know how much cheaper it is. Sure, there's R&D to integrate that with everything, but that cost is split across all units sold. It feels like the actual sensors, at this scale, can't add a significant amount to the final price.
Exactly, my previous car (BMW) once saved me in the fog by emergency braking for something I wasn't able to see yet. My current car (Tesla) shuts down almost all safety features when the camera's can't see anything, so I doubt it will help me in such situations. The only time my Tesla works well is in perfect conditions, but I don't live in California.
But "all self driving cars" are practically only from waymo. Level 4 Autonomy is the point at which it's not required that a human can intercede at any moment, and as such has to be actively paying attention and be sober.
Tesla is not there yet.
On the other hand, this is an active attack against the technology.
Mirrors or any super-absorber (possibly vantablack or similar) would fuck up LIDAR. Which is a good reason for diversifying the Sensors.
On the other hand I can understand Tesla going "Humans use visible light only, in principle that has to be sufficient for a self driving car as well", because, in principle I agree. In practice... well, while this seems much more click-bait than an actual issue for a self-driving taxi, diversifying your Input chain makes a lot of sense in my book. On the other hand, if it would cost me 20k more down the road, and Cameras would reach the same safety, I'd be a bit pissed.
vantablack
Lidar won't be effected by vantablack out side of a lab experiment. It picks up contamination very quickly and can't be effectively cleaned.
On the other hand I can understand Tesla going “Humans use visible light only, in principle that has to be sufficient for a self driving car as well”, because, in principle I agree.
The whole idea is they should be safer than us at driving. It only takes fog (or a painted wall) to conclude that won't be achieved with cameras only.
On the other hand, if it would cost me 20k more down the road, and Cameras would reach the same safety,
You had a lot of hands in this paragraph. 😀
I'm exceptionally doubtful that the related costs were anywhere near this number, and it's inconceivable to me that cameras only could ever be as safe as having a variety of inputs.
Musk's ethos is clear, both in business and government. He will make whatever short term decisions his greed and the ketamine tell him to make, and fuck whatever happens down the road. Let's not work so hard to sanewash him like the media has Trump.
Most non Tesla brands that have some sort of self-driving functionality use lidar and/or radar. I've got a BMW iX and as far as I know it uses cameras, radar, lidar, and ultrasonic sensors.
I remember reading that tesla only uses cameras for it's self driving. My 2018 Honda uses radar for the adaptive cruise so the technology exists, musk is just an idiot.
Radar would not detect a Styrofoam wall either the return from Styrofoam is extremely low. Radar also can not distinguish elevation differences very well so an overhead road sign can be mistaken for a stopped vehicle or a stopped vehicle mistaken for an overhead road sign.
Honestly all the fails with the kid dummy were a way bigger deal than the wall test. The kid ones will happen a hundred times more than the wall scenario.
Some sort of radar or lidar should 100% be required on autonomous cars.
I fully agree, but sadly, investors likely care more about their cars hitting walls than hitting kids. Killing a kid or pedestrian in the US is often a very cheap fine. When my uncle was run over on a sidewalk next to his son, the police ruled it an accident and the city refused to do anything. Same thing happened when my friend was ran over in a bike lane.... So killing humans is probably cheaper than hitting a wall.
Anyone with half a brain could tell you plain cameras is a non-starter. This is nearly a Juicero level blunder. Tesla is not a serious car company nor tech company. If markets were rational it would have been the end for Tesla.
Notably, roomba vacuum cleaners use cameras instead of lidar that other robot vacuums use. I bought a high end roomba a couple months ago and it was crap at navigating my home, while my old xiaomi with a lidar works perfectly fine. Needless to say i returned the roomba.
Austin should just pull the permits until all the taxis have lidar installed and tested. Or write a bill that fines the manufacturer $100 billion for any self driving car that kills a person and puts the proceeds 50% to the family and 50% to infrastructure. One of the first rules of robotics was always about not harming humans.
I love that one of the largest YouTubers is the one that did this. Surely, somebody near our federal government will throw a hissy fit if he hears about this but Mark’s audience is ginormous
The rain test was far more concerning because it's much more realistic of a scenario. Both a normal person and the lidar would've seen the kid and stopped, but the cameras and image processing just isn't good enough to make out a person in the rain. That's bad. The test portrays it as a person in the middle of a straight road, but I don't see why the same thing wouldn't happen at a crosswalk or other place where pedestrians are often in the path of a vehicle. If an autonomous system cannot make out pedestrians in the rain reliably, that alone should be enough to prevent these vehicles from being legal.
The question there would be does Austin have crosswalks that don't have red lights. Many places put a light at every cross walk, but not all. Most beaches don't have them at every crosswalk, they just have laws that if someone is in or entering the crosswalk you have to stop for the pedestrians. (They would all be at risk from what you are saying).
I don't know the answer to your question, but I'll add that I've seen major cities that have overhead yellow flashing light boxes that mean "you must stop if there is a pedestrian crossing the road"
Not every pedestrian follows the rules of the lights though. And not every pedestrian makes it across the road in time before the light changes colors from red to green.
This is like the crash on a San Francisco bridge that happened because of a Tesla that went into a tunnel and it wasn’t sure what to do since it went from bright daylight to darkness. In this case the Tesla just suddenly merged lanes and then immediately stopped and caused a multi car pile up.
There's a very simple solution to autonomous driving vehicles plowing into walls, cars, or people:
Congress will pass a law that makes NOBODY liable -- as long as a human wasn't involved in the decision making process during the incident.
This will be backed by car makers, software providers, and insurance companies, who will lobby hard for it. After all, no SINGLE person or company made the decision to swerve into oncoming traffic. Surely they can't be held liable. 🤷🏻♂️
Once that happens, Level 4 driving will come standard and likely be the default mode on most cars. Best of luck everyone else!
There is no way insurance companies would go for that. What is far more likely is that policies simply wont cover accidents due to autonomous systems. Im honeslty surprised they wouls cover them now.
If it's a feature of a car when you bought it and the insurance company insured the car then anything the car does by design must be covered. The only way an insurance company will get out of this is by making the insured sign a statement that if they use the feature it makes their policy void, the same way they can with rideshare apps if you don't disclose that you are driving for a rideshare. They also can refuse to insure unless the feature is disabled. I can see in the future insurance companies demanding features be disabled before insuring them. They could say that the giant screens blank or the displayed content be simplified while in motion too.
What is far more likely is that policies simply wont cover accidents due to autonomous systems.
If the risk is that insurance companies won't pay for accidents and put people on the hook for hundreds of thousands of dollars in medical bills, then people won't use autonomous systems.
This cannot go both ways. Either car makers are legally responsible for their AI systems, or insurance companies are legally responsible to pay for those damages. Somebody has to foot the bill, and if it's the general public, they will avoid the risk.
Not sure how it plays for Tesla, but for Waymo, their accidents per mile driven are WAY below non-automation. Insurance companies would LOVE to charge a surplus for automated driving insurance while paying out less incidents.
Kids already have experience playing hopscotch, so we can just have them jump between the rooves of moving cars in order to cross the street! It will be so much more efficient, and they can pretend that they are action heroes. The ones who survive will make for great athletes too.
Somebody with better animation skills than me make a cartoon where Wile E. Coyote is hunting cybertrucks using his old tricks and every single one of them works in his favor.
That's the best part, they kinda can't.
There are videos from before they pulled the sensors of some pretty cool stuff where teslas slammed the breaks before anything visibly happened, based on lidar sensors sensing trouble a couple cars up the road, completely blocked to vision.
super cool safety tech, and then they pulled it....
The thing is, RADAR can see things humans can't. There was a whole article a while back about a Model X that avoided an otherwise unavoidable accident by bouncing radar under the car in front of it and seeing that car slam on the brakes.
I’m kinda confident that even RADAR + cameras was good enough, but they started shipping cars without it and even shutting off the RADAR in existing cars.
The main negative about LiDAR is the cost, but that’s quickly going down.
I tried watching it and it forces a horrible dubbing over it so I didn't want to watch it. Apparently only way to chage it is to change my whole youtube account language
I saw the video pop up in my Youtube recommended, but didn't bother watching because I just assumed that any cars tested would be using LIDAR and thus would ignore the fake road just fine. I had no idea Tesla a) was still using basic cameras for this and b) actually had sophisticated enough "self driving" capabilities that this could be tested on them safely.
They are not still using cameras but removed LIDAR and radar from their cars during the chip shortage 2020/21.
The story they were telling was "humans don't have LIDAR but can drive cars as well, so the cars also only need 'eyes' like humans".
Small correction here: they never had LIDAR. Cars with LIDAR have big racks on top with a spinny thing measuring the surroundings. Teslas had radar but removed during the chip shortage (and disabled it on existing cars) and acted like it was an improvement. The radar was used for distance keeping on cars and could actually detect the car in front of the car by bouncing signals off the ground, it was really slick.
I'll add that every other self driving car company has a pretty good safety record, specifically because they do use LIDAR and RADAR so they can see better than humans.
That statement of him is not entirely wrong. But we humans have a very powerful bio computer that is perfectly tuned to process those visual inputs in realtime. Until a comparable performance is possible, removing LIDAR is very stupid.
They tested a LiDAR rigged car, and it stopped just like you predicted. As of 2021, Tesla uses only cameras for FSD, and not even radar (which my stupid fine Toyota truck has).
They tested the idea safely by building the wall out of styrofoam, or at least that's what it looks like when it blows apart :)
Front-facing radar is the bare minimum needed to pass the test given (fake-road wall). Many vehicles use it for adaptive cruise control, and radar is even faster than either cameras or lidar for figuring out the range to an object. 1000 Hz measuring distance to an object is enough to find both the relative velocity and the acceleration of another object. This provides enough time to apply the brakes safely when approaching a vehicle or obstacle
LIDAR is even better, and also more compute intensive and expensive to install.
I think Tesla was very short-sighted in removing radar sensors, certainly. If they hadn’t, they could’ve spent more of their energy on making the FSD cars better instead of just making them sufficiently safe with insufficient sensors
Why do you think lidar is not mature? It is radar, except it uses light and can get much more resolution than an RF radar. Or was that a joke.... That was probably a joke.. if it was then nm.
I'm surrounded by waymos every time I go out and I can tell you that if they "almost hit you" three times then that's very likely a you problem -- and a reason for more waymos, not fewer.
Anecdotally, only once have I been in a situation where I've said, "do better, robot" because I was slightly inconvenienced by it.
By the numbers there are very few waymo incidents compared to human drivers and those include waymos being hit by bicyclists who aren't paying attention.
I mean you're watching from a recorded video. I really doubt that it wouldn't be anything but obvious to actual humans eyes. i mean our depth perception alone would tell us something is wrong. You're just not watching this in 3D.
Maybe at 55 or 65 mph on a foggy day. But I doubt any person paying attention isn't seeing the obvious anchors holding the wall up and the incorrect perspective at 40 mph.
That said, other cars, with more types of sensors, would probably have "seen" the obstruction on the road.
Well yeah, that’s sort of the entire point of the video. He ran the test with a lidar-equipped vehicle, and it saw the wall right away. Hell, a radar-equipped car (like early teslas) probably would have seen the “kid” behind the wall as well. But since Musk has decided that cars should be able to self-drive with only cameras, the newer teslas will just plow straight into the wall without braking.
I would have liked to see how a more typical car with automatic cruise control /braking functioned. I think they use ultra sonic sensors and would have done better than the Tesla.
I think the human brain hasthe edge in processing visuals since out brains are so much better at adapting than any computer system. We can improvise much better since we have our whole life experience to draw on.
He is studiously apolitical, the only political comment I could find from him was the very sensible advice that we need to tone down our hyperpartisanship :)
For me, I criticize any vehicle that is objectively crappy... and some vehicles where I find them subjectively crappy... and I hope folks don't assume I'm doing that because of my political leanings.
The story of the disney thing as a reason for why to make a Lidar video, is a great "cover your ass" move.
No one will accuse him of doing it to hit Elmo's self driving taxi ambitions. but the timing is telling.
he could have made the video at any time, he chose to do it now.
Trump will never be in it. His security detail could never allow him to drive in it. Nor would the back seat ever be as comfortable as what he usually drives in so he wouldn't want to be in it unless it was for press purposes. There's a reason the Cadillac cost over a million dollars that the presidents usually drive around in.
In silicon valley there is an episode where a bunch of phones explode because of a software problem. A lot like the pager attack trump got a trophy for. And musk could take any of these cars and "self drive" them to where ever, and "update" their discharge parameters or something, then boom. The trucks are 10k lbs too. Bet you could take a small building down with one without much fuss. They are pretty fast. Scary shit. Musk is a huge problem. Watch all gov envoys being his swasticars and then he can take people out russian style. opps, accident, again.
The purpose of the video is to test a hypothesis, not to total a car.
Mark Rober is a youtuber sure, and some of the stuff he does is to feed the algorithm. But he's also an engineer, and that involves experimentation and a good dose of science.
Engineers won't set up tests that intentionally destroy their expensive test equipment if they can conduct an equivalent test non-destructively.
I’m a bit disappointed they painted identical to the actual road. Probably a lot of humans will get fooled by that one. We should send a challenge back: how looney toons can you get? Will something more cartoonish fool it? Will a different landscape fool it? How about drawing an oncoming train?
Why not? Seems fitting an ex NASA engineer show Elon, the man currently trying to dismantle NASA, just what kind of intelligent people exist in that agency.
Can this be solved with just cameras, or would this need additional hardware? I know they removed LIDAR, but thought that would only be effective short range, and would not be too helpful at 65 km/h.
If for some bizarre reason you would want to stick to cameras only, you could use 2 cameras and calculate the distance to various points based on the difference between the images.
Thats called stereoscopy and is precisely what gives our brains depth perception.
The issue is that this process is expensive computationally so I'd guess that it would be cheaper to go back to lidar.
Theoretically, yes. A human would be smart enough not to drive right into a painted wall, using only their eyeballs combined with their intelligence and sense of self-preservation. A smart enough vision system should be able to do the same.
Using something like LIDAR to directly sense obstacles would a lot more practical and reliable. LIDAR certainly has enough distance (airplanes use it too), though I don't know about the systems Tesla used specifically.
LIDAR certainly has enough distance (airplanes use it too)
As I understand it, this is uncommon and mostly used for topological mapping.
Most commercial aircraft use a radar, augmented with a GPS-based terrain map, for their ground proximity warning (EGPWS, “Enhanced Ground Proximity Warning System”).
Good question. I don’t know if they ll succeed but they have a point that humans do it with just vision so why can’t ai do at least as well? We’ll see. I’m happy someone is trying a different approach. Maybe lidar is necessary, but until someone succeeds we won’t know the best approach, so let’s be happy there’s at least one competing attempt
I gave it a try once and it was pretty amazing, but clearly not ready. Tesla is fantastic at “normal” driving, but the trial gave me a real appreciation how driving is all edge cases. At this point I’m no longer confident that anyone will solve the problem adequately for general use.
Plus there will be accidents. No matter how optimistic you may be, it will never be perfect. Are they ready for the liability and reputation hit? Can any company survive that, even if they are demonstrably better than human?
If the roadrunner cartoons taught me anything, it's that no matter what happens, whether you've been blown to smithereens or fell off a cliff and subsequently flattened by a boulder, you'll be good as now in the next scene.
Many people tend to doze off so much they would absolutely get fooled. I admit I might, too, especially if the wall is made of a material that needs no guy wires to prop it up. They either used digital effects or a very good color grading job, it's uncanny.
The biggest problem will always be a backdoor that allows remote control of the car for purposes of killing the driver or other people. The Wile E Coyote attack is much more expensive and puts attacker in jeopardy for the time involved in constructing the "trap".