The Challenge of Building a Self-Driving Car

Be one of the first 500 people to sign up with this link and get 20% off your subscription with Brilliant.org! brilliant.org/realengineering/
New vlog channel: / @brianmcmanus
Patreon:
www.patreon.com/user?u=282505...
Facebook:
/ realengineering1
Instagram:
/ brianjamesmcmanus
Twitter:
/ thebrianmcmanus
Discord:
/ discord
Get your Real Engineering shirts at: standard.tv/collections/real-...
Credits:
Writer/Narrator: Brian McManus
Editor: Stephanie Sammann (www.stephanie-sammann.com/)
Animator: Mike Ridolfi (www.moboxgraphics.com/)
Sound: Graham Haerther (haerther.net/)
Thumbnail: Simon Buckmaster / forgottentowel
References:
[1] www.technologyreview.com/s/61...
[2] www.bloomberg.com/news/articl...
[3] www.ntsb.gov/investigations/A...
[4] Tesla Conference Live Stream • Tesla Autonomy Day
[5] www.tesla.com/en_IE/autopilot...
[6] MIT Course • MIT Self-Driving Cars ...
[7] www.theverge.com/2019/3/6/182...
[8] / introducing-waymos-sui...
[9] www.spacex.com/press/2012/12/...
[10] www.cs.cmu.edu/~tom7/mario/mar...
[11] www.nytimes.com/2015/09/02/te...
[12] • Paris streets in the e...
[13] thehill.com/policy/technology...
Music by Epidemic Sound: epidemicsound.com/creator
Songs:
Sunrise In Sacramento 3 - Niclas Gustavsson
Ambivalent Thoughts - Magnus Ringblom
Zone 13 - Niclas Gustavsson
Waltzer - Dye O
Emoticon 17 - Niclas Gustavsson
The Sum Of All Things - Gavin Luke
Thank you to my patreon supporters: Adam Flohr, Henning Basma, Hank Green, William Leu, Tristan Edwards, Ken Coltan, Andrew McCorkell, Ian Dundore, John & Becki Johnston. Nevin Spoljaric, Jason Clark, Devin Rathbun, Thomas Barth, Paulo Toyosi Toda Nishimura

[@thesteaksaignant]

The "humans don't follow the rules" issue can even become a "humans see a waymo and go first because the waymo will not challenge them" Just like driving-school cars, people don't want to be stuck behind it so they tend to be more aggressive. Therefore I think tesla has the advantage of collecting "real life" data because other drivers don't know if the tesla is self driving or not

[@brytonmassie]

Sad, but true...

[@WillN2Go1]

Good point. As a recently re-newed cyclist what I've noticed suggests that reducing the issue to the AI car needs to be a bit more aggressive doesn't cover it. It's basic game theory. A lot of drivers who arrive before me at a 4 way stop won't proceed even when it's clear that I'm stopping, a lot of time they wave me through. I don't often come to a complete stop unless there're cars or pedestrians. And on T-intersections with traffic signals, when my route crosses the 't' and I won't encounter any cross traffic, I usually slow, check for pedestrians and turning cars and then proceed. More than a few times I've done this right in front of a police car. A lot of what I do on my bicycle, I'd never do in my car. The officers don't seem the least bit interested. (My theory is that it's better to get out in front of the scrum before the light turns green. This makes it easier for drivers to safely pass me moving at a steady speed, instead of me in the scrum wavering a bit as I start off.) It's basic game theory. I would argue that it's always been game theory, not just aggressiveness, or stupidity. What are the current biggest risks in driving? Drunk and drugs driving, distraction from smartphones and speed. Keep these in check and you have a driver who is paying attention, looking for, anticipating and avoiding risks they can be a pissed off youth who passes by and curses you or his sedate granny. ('Somebody ought to run you over' and 'I need to be extra careful' are both ways of I am not going to hit you.) The 4-way stop problem we used to call the Chip-n-Dale problem, they were two cartoon chipmunks who were overly polite with each other, always insisting on waiting for the other to go through a door first, "No, you go, I insist," "Oh, I just couldn't..." until Bugs Bunny or Daffy Duck barrels through. The problem in the fatal Phoenix accident doesn't need to be solved by software or hardware (though I'm sure both will improve), all the software really needs to do is recognize when it's confused or uncertain then, while nudging the driver to take over, it slows down or stops the vehicle. Which is what drivers are doing at a 4-Way Chip-n-Dale stop. It doesn't matter if one driver cheats, or is more aggressive-- it only matters not to have an accident. And I agree Tesla's huge amount of real data is its major advantage. It's a bit like us. After 20+ years of driving have you ever looked at a car up ahead and immediately knew, 'that guy is going to do something reckless' ? Sure if it's hot rod swerving, but how about the times you can't even tell what made you think this? That's the result of deep HI-L, Human Intelligence Learning-- through experience. (I wonder if Tesla's AI has noticed that when another driver is touching the left side of their face, their left side blind spot now extends almost to the front of their car? When I see this, I get in front or fall back. They're usually too preoccupied to change lanes, but if they do....)

[@Sylfa]

There's at least 2 cases where a google car got into an accident purely because the human drivers were pissed off at it. In one case the mayor of the town had allowed the cars, but stipulated they couldn't drive faster than 30mph (or whatever it was), so when going down a road that was marked for 50mph the driver behind got angry, overtook, then slammed their breaks resulting in them being rear ended. The thing is though, the best road network would be one with 0 human drivers, if all cars have to be selfdriving in a country you could have the cars cooperate to make traffic faster and safer. It still needs to be able to deal with bikes, animals and pedestrians so training them like they are now is fine, but the goal should be 0 traffic fatality, and that precludes human drivers. I like to ask anyone that think human drivers are better than AI drivers to watch some videos from this youtuber: kzbin.info Or any other dash cam videos.

[@WillN2Go1]

​@@Sylfa This is a good comment. I guess while driving and self-driving cars are striving to be safe, there are idiots just as diligently striving in other directions. When I was teaching students and I were chatting about self driving cars. A surprising number of young people don't want to learn how to drive. I suggested that if they ever thought they even might like to drive a car they'd better get on it. Self-Driving may even become illegal. An officer pulls you over, "Sir, were you attempting to steer the wheel?" "Ah, no officer, I wouldn't never do such a thing." "Where are you going?" "I'm drunk and I want to drive home and sleep it off." "That's perfectly alright, sir, but attempting to drive a motor vehicle on any street is as you know, illegal." I look forward to the day I can take long road trips and long naps or do other things while the car or van drives. I can see road trips where instead of camping the car drives some less interesting section while I'm sleeping, or cooking dinner, or writing. It would be like taking a sleeper train, or a sleeper bus.

[@PeteS_1994]

Man they need to employ psychologists to make self driving cars understand how humans drive more.

[@PracticalEngineeringChannel]

The four-way stop is such a cool example where you really have to test out the other drivers and respond accordingly. Never thought about how complicated a process that is. Merging in congestion seems like it would also be really hard to code - or at least something I would be nervous about letting my car do for me ;) Awesome summary.

[@robertpalumbo9089]

Hello there

[@PvPsFinests]

Hello there

[@Greg-yu4ij]

The only real way to solve for real world scenarios like this is to train for what a real human would do. Then you can compare your rules based heuristics against your human training based heuristics and if they match go ahead and take the action indicated otherwise, find a way to break out of the scenario that you have gotten yourself into, sort of like how the Tetris program use the pause button to break out. the alternative may be to slow down, veer slightly right or slightly left and test for a reaction from the other driver, etc.

[@Greg-yu4ij]

Note: you need to be able to make a lot more than 100 decisions per second to be able to model several scenarios at once, which is why Tesla wisely decided to up the processing power. In fact if I had to guess, one processor could be generating a model while the other processor generates the reaction to that model sort of how current deep fakes use one network to generate the fake and another network evaluates the authenticity of the fake

[@paki268]

if(goingToCrashIntoEachOther) { dont(); }

[@TheGrundigg]

nice.

[@uwunora]

Brilliant code, you should start a company

[@sneh269]

Mind blowing

[@fearofchicke]

Patent it before Tesla sees this.

[@revolver265]

Hehe (This is a joke. We all understand this isn't how simple the code is. Don't get wooooshed)

[@USSAnimeNCC-]

The minutes he said Tesla Instantly what came to mind is GCP Grey video on his trip on the loneliness road

[@RealEngineering]

Grey can make anything interesting.

[@USSAnimeNCC-]

@@RealEngineering So do you 👍

[@OKANGUVEN99]

@@USSAnimeNCC- I agree with you on both of your comments.

[@tomam258]

@@USSAnimeNCC- true

[@Psyberify]

It was impressive to see him go so far out of his comfort zone.

[@MostlyPennyCat]

As a software engineer, the software driving that car that ran over the person was embarrassing poor. You _start_ with the premise, "if in doubt, alert the driver and slow down" As soon as an anomaly is detected, you drop to that.

[@LostieTrekieTechie]

Indeed, you must prove it's safe, not fail to prove it isn't.

[@joonhshin]

Then you run into a problem of human operators eventually ignoring the alarm as the become used to too many false alarms. It's a fine balance to maintain.

[@Chilukar]

There were all sorts of problems in that collision. They had too many false emergency stops so they turned it off (I don't know why they turned off the alarms as well). They also got rid of seperate data input and driver roles, meaning drivers had to do the mutually exclusive tasks of being alert for emergencies and inputting data into a computer. I don't know if the driver in this case was "doing her job" or texting, but either way the priority should have been driving.

[@g00rb4u]

@@joonhshin Yeah in productio mode sure.. but when developing, best (especially when lives are at stake) to err on the side of caution and flag as many alarms as possible, so that they can be addressed.

[@volvo09]

@@Chilukar she was watching a tv show on her phone. Which sort of tells me uber gave these drivers pretty boring tasks, but shutting off alerts to drivers and expecting the test driver to be attentive in a mostly reactionless duty is asking for trouble.

[@MrYeezy77]

Tesla: so what feature should we add to our cars? Elon: Yes

[@revolver265]

Elon: All the things! (Jesus that is an old meme)

[@FunScientifix]

I love these YES memes.

[@atranas6018]

Elon: But no LIDAR please...

[@teemum.9023]

You 4chan men don't know what a meme is.

[@frozenthorn9619]

@CLP its a little more complicated than that, and certainly not limited to Tesla. Volvo has the same problem, in fact all cars with emergency assisted braking have this problem. The problem scenario is when there's a car in front of you that leaves the lane (because it saw what was up head) and you're left too little time to react also. This happens a lot with humans too, I saw it actually just 2 days ago, someone ahead darted quickly into my lane to avoid a stopped car in traffic, and the person behind him couldn't see or tell what he was doing in time to not hit the car. The reality is the human driver would have had a high statistical chance of hitting the parked fire truck also In terms of why the self-driving and emergency assisted braking cars can't yet resolve this is pretty straight forward. The technology used, radar isn't especially good at detected stationary, non-moving objects. Lidar is somewhat better but much more expensive, and it's not actually practical yet in terms of durability, Uber is trying to use it but it won't survive potholes and harsh weather so you have a system that is overly costly to start with that will need regular repair/replacement. Unless something can be done to make it even cheaper than it has come down to, it's still not practical for the need yet. The future is going to be a mix of cameras used in conjunction with each other for stereoscopic vision mixed with radar so you can have the best of both worlds. This takes a lot of processing power to analyze in real time so it can be useful for reacting in the amount of time that is actually needed (our brains are crazy fast computers) and thankfully Tesla already has a new processing chip exactly for this, it wasn't in those cars you've heard of in the news. BUT with those new chips, you will hopefully see fewer cars in the news doing that kind of thing, it's a process though, and Tesla seems up for the challenge, as long as we understand that all advancements take time and aren't so quick to throw blame.

[@aemmelpear5788]

"That's the beautiful thing about software. It's easily updatable" * looks at own code * * laughs in bad programming *

[@Jimmy_Jones]

Run: Programming bad laughs.

[@billcichoke2534]

That's the sad thing about programming...it's as prone to human error as other human decisions. It also happens at 100x their speed.

[@thelvadam2884]

@@billcichoke2534 Yet an AI does way less irational decisions than a human, therefore is a safer option.

[@billcichoke2534]

@@thelvadam2884 No such yet. Everything we're talking about right now is a DATABASE system, human controlled and edited. It is, therefore, only as good as the humans editing the database. Still very flawed and irrational.

[@clayz1]

That software is easily updatable is the BAD thing about it. How many times have we seen unstable updates? Hundreds of times? Software is never done, not because it doesn’t work, but because programmers just gotta program. Programmers are ALWAYS fucking with programs. When you buy a tool, like a screwdriver, you can use it for a lifetime, or tell just by looking that it is wearing and needs replacement. Users don’t ever know what surprises are built into the latest release, and the same is true of the programmers who built it. Thats the reason why we have weekly, daily or hourly “updates”. They are mostly just bug fixes masquerading as “engineering updates”.

[@aurobhatta]

The programmer probably did it like this While something not identified: Run over it If something identified: Break

[@Tiwi]

Self driving vehicles's safety can't be measured by death/mile since there aren't enough fatal incidents to have significant statistic. As Elon said in a recent interview talking about Autopilot it's far better to measure the safety using a incidents/mile indicator.

[@zephyr6877]

No. It just makes self driving cars look better this way because they do in fact lower the chance of very minor bumps/collisions which usually happen from human error. but the accidents like this that really matter because someone died shows that self driving cars are far from ready.

[@mjc0961]

No, it's not better to measure using a statistic that Elon cherry picked to make his cars look good.

[@Tiwi]

This is the first fatal incident happened in a full self driving car, good luck building a statistic death to mile.

[@Tiwi]

@@zephyr6877 Do you have any proof of what you are saying? I've seen the compilation of near incidents not happened thanks to autopilot. If you say that autopilot is mainly useful only in the avoidance of minor bumps you have to prove it.

[@GURken]

What about another challenge - infrastructure? Potholes, barely visible road markings, flooding because of clogged storm drain, uncleared snow etc.

[@dracoeris]

Aka Driving in Canada

[@benheinz8817]

@@dracoeris I was in Pennsylvania just a bit ago and their roads are horrid.

[@aethersgeckogaming2176]

I might be wrong but I think they have a thing where they can predict what the most likely path would be and use that to guess. I'm probably wrong but I've seen a video of one driving on an unmarked road.

[@dirtybongwater5751]

@@aleksandarilievski7273 Where I live we regularly get above 37 C in the summer

[@cerebralm]

Well, how do humans solve that problem? The Ai isn't checking for clear lane markings, its analyzing all the available data to find any patterns that may be present. If it can plot a safe course using only the behaviour of other vehicles (the way humans guess at lanes in snowy areas) then it will do that. Tesla has access to enough data from enough edge cases to learn a behaviour for those scenarios.

[@JS-kv8ey]

AI: Why do accidents happen Lmao just stop driving. *Accident rates drop to zero*

[@naufalap]

natural selection: allow me to introduce myself

[@prasadvatsal]

we did to boys, accident is no more

[@typie34]

Harvard: Nigga u want a scholarship

[@zachdurocher1166]

"Make everything legal" *crime rate drops to zero*

[@billcichoke2534]

@@zachdurocher1166 Trying to redefine criminal acts as noncriminal does not make them any less destructive. Anarchy is criminals not wanting to pay a consequence for their rebellion. Might makes right is what results. Careful what you wish for...there's always a bigger bully in the pond with you.

[@wompwomp9904]

I saw a video on pornhub with a auto pilot tesla and its advantages. Great vid btw awesome demonstration, he had a girl with him in the car i forgot what they were doing but was amazed by the tech.

[@dirtybongwater5751]

The same video got recommended to me

[@McDoodle44]

Link please 😁

[@creativemindplay]

:P :)

[@blue_leader_5756]

That is both hilariously awful and awfully hilarious

[@MrEmrys24]

I saw that too. Her bobs and vagen was shown so probably they were doing physical exam.

[@nickgehr6916]

The challenge of building self driving cars: *human stupidity*

[@Yamyatos]

I just had an urge to click human stupidity because of the positioning and bold letters lol. Also i dont get why people get angry at his comment. The main problem is actually humans not following rules. Sure.. the car killed that person, but only because the person ran in front of it. It's true that we need to programm / train it in a way that allows to handle such situations better while not decreasing over all speed in other situations (which is hard), but looking at how impossible it is to see anything at 1:48, most humans would have killed that person as well. In a couple decades self driving technology will have improved to a point where it will become illegal to normally drive a car because of safety concerns. Which will be a good thing^^

[@lockhart1895]

The human poorly programmed the system

[@Jimmy_Jones]

Or humans fucking.

[@tungstenwall474]

@Zenn Lozanno You are not wrong. kzbin.info/www/bejne/jJi0qHt6e9SbitE Some morons assume humans will never have to touch their steering wheel again. Now he is a pancake.

[@gunarsmiezis9321]

Just be prepared for the day Where; Using car denied. reason - you are a dissidant.

[@asanezz1111]

One of the main challenges is actually stopping people from filming porn on Self-driving cars Oh wait...

[@sneh269]

Lol

[@leoaw2377]

Fake taxi using teslas 😆

[@Azakadune]

tru.

[@LinksSpaceProgram]

Inb4 someone posts the link to it lol

[@nimbusws5946]

Then they install some fucking machines in there too

[@jborynec]

Work with the glass cockpit shows that relying on pilots (i.e. drivers) to intervene when automation exceeds its boundaries is extremely unsafe. The time it takes for the driver to gain situational awareness is often way too long for whatever bad things is happening. It is much safer to keep the driver as a driver, rather than have them be a "manager" and only be a driver in extremis.

[@OfficialFront]

just a matter of the automation level

[@jborynec]

@@OfficialFront Has more to do with requiring pilot/driver to continuously maintain awareness, rather than need to rebuild it in a time-limited situation. Thus the type of automation is key, rather than the overall level.

[@jeffbenton6183]

The reason why cars crash isn't because drivers are human, it's because they're unprofessional. If there was more of a culture of driving like you life actually depends on it (like bus drivers and airline pilots) then we'd have fewer accidents. All the sensors and computers are great if they augment the drivers ability (stay in the lane, watch for blind spots, etc.) We start running into problems when its specifically designed to make the driver feel unnecessary and stop paying attention.

[@jborynec]

@@jeffbenton6183 Are you suggesting that Airline pilots are unprofessional? The reason that we know about this type of human failure is because airplane safety is extensively studied.

[@jeffbenton6183]

@@jborynec I thought I was saying the opposite. Flying is safe because pilots are professional and every incident is closely studied. We don't do that with driving.

[@elyakimlev]

LiDAR has more disadvantages than just the cost. It doesn't give you the meaning of objects in the scene: It can detect a shape on the road but it can't tell you whether it's a bag or a tire. It can detect a human but it can't tell you if the human is just standing watching their phone or intently looking to see if it's safe to cross the road. It can detect road signs but can't read them. All of these things can be solved with cameras if you have big enough data sets and compute power.

[@peepiepo]

A camera doesn't tell you what objects are either

[@elyakimlev]

@@peepiepo You're right. The radar complements that. Camera + radar beats LiDAR in every way.

[@Steppenkater]

Elon also showed that the computer in a Tesla is powerful enough to calculate a 3D-environment from camera-data alone, getting all information a LiDAR would provide. So LiDAR is only good at the beginning but high-perfomance computer and cameras will overtake LiDAR soon or have already.

[@MarcusHast]

@@elyakimlev No. Radar also doesn't give you any knowledge. Computer vision does that. And it works on LIDAR as well. Sensor fusion is also not limited to radar and cameras, there's nothing stopping you from fusing LIDAR information into the system as well. (Well, besides cost.)

[@Rishab12312]

Lol u are just wrong

[@elliot7452]

Having watched the autonomy day presentation and one by mit I wasn't sure if I would learn much from this video but was pleasantly proven wrong. Great information and perspective

[@JonTingvold]

Second that.

[@ShanaXinator]

Amazing work compiling this video. Love it! Also joined your Patreon too! Keep up the good work my man :)

[@scott98390]

if (cantDetermineObstacle) { decelerateSlowlyToGainMoreTime() }

[@minecrafter0505]

Would result in quite slow cars sometimes though.

[@PKMartin]

The section from 11:53 demonstrates why this won't work - most of the time the car isn't quite sure what something in its field of view is, so it would constantly be slowing down if safety was always the only concern.

[@Yamyatos]

Yeah.. the problem is how to fill in "cantDetermineObstacle". Objects are classified using a neural network, which was trained to determine the _probability_ of something fitting into each category it was trained to recognize. The car is never sure or unsure.. it always has a gradient of "certainty" and selects the most probable option as the one to categorize the object as.

[@galihprabasidi8498]

@@PKMartin I think it can work if the car also determines whether the obstacle trajectory and car trajectory intersect

[@user-nj1qc7uc9c]

invalid syntax, missing ";"

[@zeyad45]

Super analysis! One of the best til date.

[@vallorahn]

5:35 with subtitles: *cut to footage of Tesla speeding up autonomously to avoid a crash* and no cut never comes :D

[@Rebius]

probably re-edited the video and forgot to edit the subs :D

[@MenkoDany]

LIDAR resolution is actually really low, especially compared to cameras. The information per ""pixel"" is a magnitude smaller than in a camera. I think the word you were looking for was precision?

[@weasle2904]

I mean he said "high resolution lidar" as in it's relatively high resolution for being lidar. Anyhow the purpose of the system is to gather 3D information about the surrounding area thusly being more accurate at recognition than a camera

[@tartiflette6428]

Something missing here is how self driving cars could also communicate with eachother instead of trying to guess what the other cars will do, making them _dramatically_ safer as they get more popular.

[@sircatington6793]

But that is very, very far off considering the fact that many people will be skeptical at the prospect of self driving cars and that we don't even have a fully self driving car yet, but when it does happen the reduction in traffic is going to be glorius

[@sircatington6793]

@@tartiflette6428 Yeah once self-driving cars have been around for a bit and society starts to accept them they'll definately explode in popularity and a lot of stuff related to them will change to accomadate that.

[@ronitrajput3934]

That is why Ultron was trying to protect everybody, by killing all of em, Hail hydra.

[@zachfakelastname]

Hail hydra

[@mcmarkmarkson7115]

Imagine the high tech civ Ultron would have created by wiping all the human peasants out with Ultron constantly self improving. Even Thanos wouldn't mess with that.

[@rollerskdude]

This had better be some kind of joke.

[@flybeep1661]

So the Uber accident was a human mistake anyway. The system saw it, it just wasn't programmed to act. The sensors saw it, the system was capable but it was not told to act.

[@MichaelSteeves]

That is one of the outstanding problems: Who is liable for such accidents? The driver or the programmer?

[@RedStefan]

Easy, the taxi service for using technology that is not ready to be used.

[@N7a7v7i]

The actual answer is that as always happens in real life, no one is ever liable, because our legal system does not give a single fuck about anyone killed by a motorists. Pedestrians and cyclists barely even count as second class citizens.

[@Sk1erDev]

A Tesla service manual showed the model 3 would be getting a second forward facing radar for redundancy

[@blink182bfsftw]

Will they finally stop crashing full speed into trucks stopped on the highway?

[@DB-gh4nj]

@@blink182bfsftw No as the engineers at Tesla find it too difficult to merge radar and camera data Tesla is no limit nger using radars. Can't make this shit up

[@AaronSchwarz42]

Driving is an insanely dangerous task because of the vast multitude of different situations that can come up etc The state *pick you area, country etc* failed royally to drill this into the heads of its populace, the public Today you have a mix of highly skilled safe drivers with dangerous poorly skilled drivers & mix of people of moderate skill levels as drivers = accidents 1 Deadly accident per million miles is not that bad ... in the USA ... where the roads & driving are fairly well organized ! The cool part about self driving vehicle technology is that it will just keep improving while people stay about the same as a whole, on the average ...

[@superdau]

Since all road infrastructure is built to be navigated by humans (and it will be like that for a very long time), cameras will have to be the main sensor for autonomous vehicles.

[@vishaltripathy3620]

Amazing video. That tetris story is mind blowing. I hadn't thought about that possibility. Thanks for the video

[@kevind814]

Issues with self driving vehicles not addressed here are Insurance and liability: Am I protected when my vehicle's "perfectly" followed programming causes human fatalities/injuries? Am I at fault for my vehicle's actions? Am I still the "driver", or just a passenger (can't fault the passenger)? Second, I'm curious if these vehicles are programmed to break speed limit laws? I got the impression from the video that they "go with the flow" speed-wise. Since the vast majority of traffic is moving faster than the speed limit (not counting rush hour of course), these vehicles would be knowingly speeding as well. Also, requiring humans to be paying attention enough to take over at a second's notice defeats the whole point of having a self-driving car. Passive driving (not paying attention to surroundings) is the problem we're trying to overcome; giving the driver less active involvement in the car's operation (steering, braking, lane changing, etc) just leads to more passive driving.

[@moartems5076]

Kevin D How speed limits are handled is mostly BS. Tourists in switzerland get a speeding ticket for 53 in 50, swiss policemen: "If we wanted you to drive 53, we would have a sign saying 53." Society is built on rules!

[@seraphina985]

@Yevhenii Diomidov Probably would depend whether the manufacturer had certified the vehicle for operation without an operator once they do then it probably would be liable otherwise it's likely to be like any machine that needs an operator the operator will be at least partially liable, though the manufacturer may still be partially liable if some negligence on their part contributed to the injury.

[@aritakalo8011]

@Yevhenii Diomidov It isn't up to Tesla to decide. Hurting someone goes to criminal law and courts. They decide who is liable, not Tesla. Tesla could say "we are liable", but the courts might say "no you aren't, the owner ordered the car on the road, they are liable". To begin with for the car to be legally self driving on public road, it would need government permission. Driving licenses are a thing. That car also essentially needs a "autonomous car control system driving license". When that starts to be worked out, the legal mess starts. Who is liable. If the government has certified it fit for traffic, is the government liable? The car maker? the owner? the passenger? It will be a years and years long legal mess, that might end up with "Government would be liable, government doesn't want to be liable, no self driving cars will be certified for traffic. Safety drivered cars? Sure, since then safety driver is liable and not the government" Since we are talking public roads, one can't just assume "everyone agrees to the risks of autonomous cars and accepts that the liability regime is X". Also what would be the sanctions for criminal behavior by self driving car? Can't throw the car in jail. Fines? That seems kinda insufficient, if a self driving car drives straight into a school bus and kills children. People go to jail for vehicular homicide. Does the government inspector who signed the type certification to jail? The owner of that specific car? The maker of the car? the passenger? if there is passenger to begin with anyway. Anyway is it negligence or murderous intent..... The log shows the car sees the bus, but just keeps driving. Can't exactly ask a computer algorithm about it's intent, if the date is unclear or due to some glitch it is unclear. Is random bit flip jamming the throttle decision to a latch state, negligence or intentional.

[@jazzfan1994]

TL;DR I think it would be case-by-case and probably part of a complicated regulatory framework. I suspect what would happen is an investigation (as is already common in crashes resulting in serious injury or death) to determine whether the fault is due to owner tampering, improper maintenance, or an error on the part of the manufacturer. For the last one the question becomes whether the manufacturer was obeying regulations, whether they could have been expected to know about the error, and if can be reasonably compelled to correct the error.

[@seraphina985]

@@jazzfan1994 Indeed it is probably quite likely to lead to more complex accident investigations in order to determine liability for the accidents that will happen. Still this is hardly something the authorities lack experience with they already deal with such multi factor investigations when dealing with accidents involving modern aircraft, trains etc that have complex safety systems installed and extensive maintenance histories to investigate etc.

[@CJ_McK]

Just finished watching the video, auto play took me to another of yours I'd seen before. It dawned on me how much easier it has been to understand complicated subjects like these since discovering your channel. Just wanted to say thank you for creating quality content. I appreciate you.

[@thenickle9301]

I just want to say I'm about to finish high school and wasn't quite sure what I was gonna do but your channel and other factors really have set me up with engineering. I just want to say thank you for making it interesting and getting me into the possible subject

[@SeaJay_Oceans]

You should re-title this: "The challenge of building a terrorist weapon without the suicide bomber". In the current global / political environment - Self driving car bombs can not be permitted into any major populated city, nor allowed close approach to civilian assets, government buildings, military bases, energy plants, chemical plants, factories, etc. etc. The Nightmare of self driving weapons is the Big Hack - turning millions of them into self crashing vehicles with one ''software update''. Or just stopping them all dead in the road, nationwide, world wide... dead cars blocking everyone.

[@JustinDubbs]

Hey, This video is a bit old and i feel like it might be really nice to remake this video. especially since there has been so much development in the self driving scene with lots of news, I think it would make for a really good video and would be wonderful to see how much self driving has expanded compared to what this video predicted.

[@uzairsuleman2818]

This was definitely a very impressive video, very well put together and it was easy to tell the amount of work that went into it. Keep it up man !

[@jeffk412]

I hope brilliant pays you well! 1) your content is fire. 2) your pivots to their pitch are seamless! 3) While I've seen the brilliant pitch on other content creators' sites, I decided to try it out because of the 1-2 punch you provided. Please keep up the fantastic content!

[@PixelSheep]

Man thank you for your videos!! They are so freaking brilliant! Thx

[@techman2553]

Wait until kids start pranking self driving cars by holding up stop signs, or other road signs.

[@uvmale01]

Such great detail and such a great story to wrap the concepts around.. bravo Dude.. you're the best !!

[@xomm]

In practice, radar can be pretty poor in snowy conditions as snow accumulates on the sensor. This has happened to me in pretty much every appreciable snow event I've driven in in my Model 3.

[@deelowe3]

How is this any different than a camera?

[@Pau_Pau9]

Having lidar on your self driving car is like wearing spinning propeller hat.

[@luigimafia10201]

Hehe

[@Jimmy_Jones]

Inspector gadget.

[@marieisabelvermelho152]

Being a European with US driving experience, I must say that 4 way stops are ridiculously ambiguous and stressful for humans as well.

[@frozenthorn9619]

agreed. We do actually have many roundabouts in the US too, but it's never made sense why we adopted the 4 way stop, it's ineffective and even people used to it can't seem to get it done effectively.

[@indiebekonn]

I will never understand why they’re not just “priority to the right”

[@danielrdowding]

This was one of your most engaging videos yet. Absolutely love your content!

[@ketsune23]

I found your channel today ! I am an mechanical engineering student and I dont know how to start so many good videos and so little time. Subscribed.

[@zuko1569]

"The challenge of building a self-driving car" My brain: _If it's unavoidable should the car hit 3 people that crossed the red light or hit 1 person that obeyed the traffic light to minimize the casualties_

[@geoffreychadwick9229]

Ultimately that'll be the ethics question - engineers and developers will break it down via failure analysis and risk assessment to say which carries the highest risks of fatalities/injuries. The same could be said about a pedestrian in a crosswalk (likely fatality) or swerve into a tree (risk of fatality or injury to the occupants of the vehicle). A computer having to decide who it tries to save is a legal nightmare.

[@MetallicReg]

Zuzu If you kill enough law-abiding people the remaining will have more incentive to break the laws to survive. Social Evolution 101 Ergo: Try your best but keep the main damage on the side that doesn’t follow the laws every citizen agreed to respect by being a citizen of that place. Therefore: Killing the three to rescue the one will have a smaller impact on the health of the whole society. Sorry for the three though that they haven’t learned fast enough while still alive.

[@lorddashdonalddappington2653]

Feel like the answer is just "engage brakes" since swerving is always dangerous. Come what may past that point.

[@AaronShenghao]

The reality is an self-driving car will never end up in that kind of situation...

[@revolver265]

@@lorddashdonalddappington2653 What if it's too late? There's a website I believe where it's about which group of people you need to hit if you know what all of their jobs are or endangering occupants inside by swerving into a barrier or tree to stop. Edit: Moral machines [moralmachine.mit.edu]

[@wv_]

One item overlooked is lidar is computationally extremely expensive as the RGBXYZ point Cloud it creates requires a lot of calculation with each point being compared to its surrounding points to build a 3D model. It is also laggy at a fraction of the frame rate of a camera, not practical to stream or easily store and review and point clouds are less compressible and use up huge amounts of space. Another issue is lidar is much more accurate at recording colour compared to cameras which have lots of issues with lenses and chromatic inaccuracies.

[@robieb0i]

LiDAR doesn't give colour. it gives a Intensity value, often encoded as a number between 1-256.

[@Hesperideskeeper]

Nicely composed overview of the self-driving car revolution that is currently under way. Prophetically, your points specific to current limitations and the future are insightful. A review of the history of technology and its integration into society is instructive in this regard. In all cases where the two have managed to coexist successfully over time, not only has the technology been “improved”, but importantly, we, the benefactors, and our way of living life has required change, often profound in nature. I look forward with fascination to the technological/sociological pull and tug between the conflicting requirements of safety (zero relative motion) and efficiency (maximum relative motion). Cheers!

[@Chaoticmongers]

Actually, the main sticking point against LiDAR isn't cost, it's sensor information. Cameras give so much more information about surroundings, and can tell the difference between a pedestrian and a car much easier than a LiDAR can. With a sufficiently powerful nueral net net, and enough data, self driving is possible with only cameras, with the radar acting as a safety net for redundancy. After all, human drivers rely mainly on our cameras, our eyes.

[@welyum7308]

And nobody would buy a car with this thing on the top

[@MrClaytonskaggs]

Your diagram of the uber Accident is off by at least 30 feet, you can see the pedestrian was impacted a few feet before the solid line switches to dotted. If you notice from the satellite view there appears to be an "X" shaped sidewalk in the median, this is purely a decorative (stone pavers) feature but appears to be a sidewalk. There where multiple signs telling pedestrians to not cross here as this is not a sidewalk/walkway but after the accident this feature was removed and replaced with gravel to further discourage crossing. Phoenix has one the highest pedestrian/cyclist collision and death rates of any major city in the United States. Though we are trying to understand why this occurs here more than other places the role of the design of our roads/infrastructure is one of the major causes, also crossing mid block is a contributing factor to a majority of the deaths here in Phoenix. Like many accidents it's easy to blame one party but the reality is multiple failures (Uber's Systems, Mid block crossing in the dark, poor infrastructure design) combined to create this fatal accident. As someone who rides their bike through this intersection multiple times a week and had a close call at dusk with a pedestrian crossing in the same spot 3 months prior to the fatal collision I believe it's naive to place the blame solely on Uber.

[@TimothyWhiteheadzm]

The Uber situation was reminiscent of many software projects where 'let it break then we will fix it' is the norm. Unfortunately that is a bad strategy for self driving. I think authorities should do basic tests on all self driving cars before allowing them on the road. The simple 'pedestrian crosses the road' test should be the first. Something you didn't mention is compute power. It is essential fro AI to have very large compute resources for the company to train the AI, then fast reliable compute in the car to actually use the AI for driving. For the first, Google likely has an advantage, but I do not know if Tesla is in any way limited by server side compute. In the car, Tesla claims to have developed the best in car compute - with more to come in the future. But as is always the case with computers, in 10 years, everyone will have access to similar computing power for a fraction of the cost, so Tesla will either loose its advantage or have developed even better chips to process video instead of still images. There is still room for a lot more image processing such as stereo video and higher resolution so it will be a while before computer speed is no longer a bottleneck for self driving AI.

[@jamesheartney9546]

𝘐 𝘵𝘩𝘪𝘯𝘬 𝘢𝘶𝘵𝘩𝘰𝘳𝘪𝘵𝘪𝘦𝘴 𝘴𝘩𝘰𝘶𝘭𝘥 𝘥𝘰 𝘣𝘢𝘴𝘪𝘤 𝘵𝘦𝘴𝘵𝘴 𝘰𝘯 𝘢𝘭𝘭 𝘴𝘦𝘭𝘧 𝘥𝘳𝘪𝘷𝘪𝘯𝘨 𝘤𝘢𝘳𝘴 𝘣𝘦𝘧𝘰𝘳𝘦 𝘢𝘭𝘭𝘰𝘸𝘪𝘯𝘨 𝘵𝘩𝘦𝘮 𝘰𝘯 𝘵𝘩𝘦 𝘳𝘰𝘢𝘥. This doesn't really help much. The basic tests are easy to design for, but don't tell you how the system will perform in all the edge cases. Proving that an autonomy suite is safe is a surprisingly difficult problem by itself.

[@crusherven]

@@jamesheartney9546 No kidding. Any auto company could write an AI that will pass an entire test suite. It might actually make things *worse* because companies will program to the test, and then suddenly the responsibility for the AI being inadequate is on the government agency that created the test.

[@TimothyWhiteheadzm]

@@jamesheartney9546 I fully agree that there are no good tests for highly reliable self driving. But a car that does not stop for a lone pedestrian crossing the road in a clear unobstructed view situation should NEVER be allowed on the road. Its like giving a blind person a driving licence. There ARE basic tests that would do a good job at providing a minimum standard that a self driving car should reach BEFORE even being allowed with a human driver monitoring behind the wheel. Then there should be a minimum number of miles per accident with human monitoring before the human can be taken out of the loop. But Uber failed to even pass the most basic test. This is not unlike the 737 MAX situation where a single sensor failure was enough to bring down a plane in a situation where two sensors existed and the system designer was just too lazy to think.

[@seraphina985]

@@TimothyWhiteheadzm Honestly one way I can think of to get the basics would be to pretty much run the car through multiple rounds of the human driving test under different road conditions, if the human supervising the certification has to take positive control of the vehicle due to a safety issue during the test it's a fail and it is not ready to be operating without constant supervision. After all we don't allow humans to operate vehicles unsupervised without being able to meet that test standard so doesn't seem like an unreasonable floor to start with especially since humans are generally expected to demonstrate competence in all the key areas before even taking the tests mostly so doing it in multiple rounds to ensure it works well in various road conditions seems reasonable.

[@aritakalo8011]

@@TimothyWhiteheadzm Only way really is having them out in the normal public traffic in large numbers with safety drivers for YEARS. Only way to test the unknown unknowns of real world and expecting the unexpected is to have the control system be in the real system. Not even the most sophisticated simulation can replace the unpredictability of the real world and the gargantuan variability. Snow on the roads, raining while snow on the roads. Well we just happen to have a road work, but this time a moose is running through it. Oh and there is a guy coming on the wrong side of the road and at the same time tennisball is flying past the radar at close range from kids at the next yeard, on top of that someones dog spotted the flying ball and is jumping for it from the otherside of the road. The one in the millions no one can simulate, but happen regularly, because cars drive so many miles and there is so many cars. When one in a million unexpected becomes "You must handle this regularly". Possibly even decade or two and then showing statistically, that the safety driver doesn't need to intervene anymore and that the cars have smaller incident rate than humans. MUCH smaller, since with human driver there is a liable party: Driver. It simplifies safety and regulation. If that is to be removed, it is going to be a legal mess everyt ime accident happens. So the rate has to be really small for it to make worthwhile to take away the easy liable party, driver.

[@mikecurtin9831]

Thanks very much for the information on how things currently work. Your presentation (as always) is excellent.

[@adic8993]

Amazing video man. Kudos. Keep up the great work.

[@unvergebeneid]

Wow, the description of Uber's design decisions had me speechless. I hope people end up in jail for this, both at Uber as well as the regulators who failed to regulate.

[@zachfakelastname]

To bad the American government doesn't believe in regulation till after people die and sometimes not even then

[@aritakalo8011]

Well from liability point of view, it is the safety drivers fault. From legal point of view, there was no "autonomous driving". It was a car driven by the safety driver 100% all the time, that there was self driving assist doesn't move the liability. It was allowed to road, because the safety driver had a driving license. The car had no driving license. There was nothing for regulators to regulate on the cars decision making and programming, since from their point of view it was a human driven car. Just with lot of driver assist, but still a human driven car. Thus the one who ends up in jail is the safety driver. They were criminally negligent and caused a vehicular manslaughter via their negligence. That the car was self driving and soothing them to not pay attention is irrelevant. They were 100% the driver, driver is liable for everything the car does. Not that it makes the design any better or Uber any better, but as long as there is driver behind the wheel and the accident was caused by bad driving decision and not say mechanical failure, then driver is liable and criminally liable. Hence why the safety drivers will be staying for decades to come. Because should they be removed, then actually government kinda would be liable. Since they would have to certify each self driving model. So any failure of any self driving model, would also be failure of proper certifying. As long as there is legal person with free will behind the wheel with the option of "stop the car and not drive further", the liability is clear. Driver is liable.

[@jolez_4869]

@@aritakalo8011 Driver was liable, but that collision was completely unavoidable even if there was no autonomous car and the driver was actually driving. The dude showed up in like the last second and the car was going at 70km/h. And what the fuck was he doing at the middle of the road anyway?

[@ivandiaz5791]

@@jolez_4869 The footage released to the public was absolute bullshit. The real lighting conditions on the road allowed the woman crossing the road to be seen very far away. Uber intentionally used a very poor dashcam that shows conditions being much darker than they were.

[@jolez_4869]

@@ivandiaz5791 Oh well that changes things...

[@internetisinteresting7720]

Nice video, hello from Oriente University here in Cuba. We love this channel

[@Skip2MeLou1]

Hello Oriente University

[@Jimmy_Jones]

Oriental

[@robotswithryan8348]

Your videos help me to remember why I love engineering. Thank you for all that you do.

[@Pompnumber]

Thank you for the proper explanation, makes sense to me now. So much value in this channel, thanks bro! :-)

[@LoggeL]

Tesla's weak point is that they only work during nice and clear weather. The cameras will fail when dark or in a snow/rainstorm. I feel like this problem isn't considered enough.

[@paulgarcia2887]

Infrared camera + Night vision camera Problem solved...

[@LoggeL]

@@paulgarcia2887 a NV camera doesn't help when it's snowing and raining. Also they produce a very noisy and low quality picture. They might be helpful for simple low light conditions but they won't solve the overall problem.

[@limiv5272]

Humans can't see well during a snow or rain storm either. The self driving car doesn't have to be perfect, just slightly better then a human driver, and a human dorsn't have the benefits of infrared or night vision, so that shouldn't be hard

[@LoggeL]

@@limiv5272 You are 100% correct. I'm just talking about the cases where for example the Camera gets covered an the system suddenly loses it's "vision". A fully autonomous system would have to stop and wait for clearance. There is too much to debate and there are still too many open questions about fully autonomous driving. I'm excided for the future development.

[@waynelewis9110]

Logge For some reason, some people forget that Tesla’s vehicles aren’t only relying on 8 cameras. There are 12 ultrasonic sensors as well as a forward facing radar. The ultrasonic sensors + forward facing radar permits the car to work very well in bad weather conditions, something that LIDAR is incapable of despite being super expensive.

[@gunfumaster1024]

It's funny we get self driving cars before we get "smart" traffic lights...

[@wiebemartens1030]

You probably don't live in the Netherlands, where trafficlight systems are programmed to react to the flow on traffic

[@Smt_Glaive]

Fake taxi on a new level.... Can't wait

[@Smt_Glaive]

I'll get the likes rolling

[@JotaSzeps]

Excellent video! I've learnt a lot with it! Thanks!

[@NoOne-fe3gc]

2003 FPS? But can it run Crysis though? Are we getting an announcement for Skyrim on Tesla? these are the real questions ppl should be asking

[@dirtybongwater5751]

Finally someone brave enough to ask the questions that really matter

[@Venaloid]

Honestly, every time it rains heavily, every time the lines aren't painted too well, I think, "Yeah, like a computer could figure out where the road is in these conditions."

[@MichaelSteeves]

My Subaru with eyesight does remarkable well in poor weather but it isn't perfect. I have to turn it off when driving after a storm as the worn paths often cross the lane markers and the "lane departure" alert goes off repeatedly.

[@arun3151997]

We also thought that it would be impossible for a computer to talk like a human yet here we are..

[@OKANGUVEN99]

​@@MichaelSteeves Thats kinda the problem I think. We can make cars that help human drivers to a point of making crashes 10x rarer than they are now. Getting to a point of a car driving itself %99.9 of the time is relatively easy but that %0.01... That is tough.

[@isaiahphillip4112]

That's why I personally don't really enjoy the idea of self driving cars until we have something closer to general AI. I want my car to actually be intelligent and capable of improvisation to some degree, not just a very efficient rule following drone. There are literally an infinite amount of potentially hazardous situations that could present themselves to a driver, you can't even dream of programming a response to all of them, so the ideal should be an artificially intelligent system that can (to a degree) develop work arounds and solutions to new problems on the fly.

[@jamesheartney9546]

@@OKANGUVEN99 That's called "chasing the nines" and is a well-known part of the self-driving design challenge. You can get to 99% safe, but what about 99.9%? 99.99%? etc. Every last increment becomes exponentially harder. There's a school of thought that says full vehicle autonomy will prove impossible to implement because the last few nines will be intractable.

[@Oscee613]

I work on self-driving cars for years now and few things to add: - braking the rules makes traffic most effective in most situations, this is very hard to model - if everyone speeds but you obey the law and are lot slower than everyone else, you are actually dangerous - traffic rules are different between regions, human behavior even more so. Not only that, human behavior is quite conditioned on situations (is it raining? do I have a little child in the back?) and humans can not only negotiate but influence each others' decisions (example: I'll roll just slightly forward to signal my intent and make someone let me in - see: Waymo car failing to merge). No one, NO ONE has a good representation of human behavior and negotiations skills. And won't for years. - some of this tech is quite US-centric but in my book, any decent self driving should work globally. I've driven on crazy Hong Kong streets, Japanese rural hills, Italian small town cobble stone streets; US highways are just incomparably easier to handle (let me not get into how stupid the 4-way stop is, glad it doesn't really exist outside of N-America :). This includes types of vehicles (do you detect TukTuks?), different architecture and even language (can you read a sign in Korean saying "Only school bus between 7.30-9.00am"?). Sadly people can be included in this too (the most widely used open source perception model regularly fails to detect "woman with uncovered legs" and lot of them have worse performance on people of color). - you have left out a few sensor modalities that could be useful (quite underrepresented in current available tech too): heat/infrared and sound/vibration. True, first is technically a camera but has very different use-cases. Latter one is super useful for detecting road surface, oncoming ambulance, etc. Hyperspectral seems to be an overkill for this usecase but who knows. - there is only so much more data allows you to train. There is a growing realization in the industry that our current AI/ML technologies are possibly inherently not enough (certainly very much not efficient) to solve these problems. If you've seen a reinforcement learning agent try to learn robot control, you know what I mean. Superb video though, thank you for it. One personal thing to add is I'm quite sad about the hype this tech has gotten (khm, Musk). Not only unsafe and unethical but gives false impression to people that it is ready and near. Full-self driving is very, very far away. Lot of "traditional" carmakers have great tech but they don't need to be vocal about it to keep their companies alive and inflate shares plus they have 5+ years product strategies and marketing plans with no need to hype up unfinished features.

[@Oscee613]

Disclaimer: yes, both Tesla and Waymo have amazing tech and are clear leaders of the field - if not in functionalities, certainly in productization.

[@Pixy643]

Not just Cars , we need sensors on the roads too, and self driving Cars need some form of communication with other vehicles on the road for increased safety.

[@frozenthorn9619]

Uber rushed many aspects of the testing phase, I'm sad as a programmer to see they overlooked such a basic requirement. When you don't know what's going on ahead, slow down until you do, if it still isn't clear, come to a stop. That's what a human would do, and if we're trying to build tech that is better than humans, we need to at least aspire to be as good first. The biggest challenge to automated driving is definitely going to be other drivers though. We'd be able to almost instantly cut accidents and deaths in half if we got rid of all human drivers, and used a hive network of Tesla cars, that's the future we need but it requires gradual change, and requires being able to co-exist until such a time as human drivers no longer make sense. I'd love to take a nap or watch tv on my way to work, we'll never get there without the work Tesla and a couple others are doing, so we need to support them in any way we can. Change is always hard and there's going to be an adjustment period, but we can't let it stop the progress.

[@gregcollins3404]

Autonomous cars need to have a safety standard way higher than human drivers because responsibility is being transferred from the driver to the software engineer. The liability being taken on by the car company with presumably deep pockets is going to be difficult to surmount I think. Tell me the average driver isn't going to blame any incident on the computer.

[@Im-mv6bf]

autonomous cars are already the safest cars on the road. can't prevent what other drivers do though

[@paulgarcia2887]

Meh. If the car signals the driver to take the wheel and the driver fails to take action then blame should be put on the driver. If the car hits an obstacle and it did not alert the driver prior then blame should be put on the company.

[@mvmlego1212]

@@Im-mv6bf -- Did you watch the video? He explained that there isn't enough data to reach conclusions about that, even under normal driving conditions.

[@Im-mv6bf]

@@mvmlego1212 Tesla vs same amount of ice cars.... which has more crashes ? the answer is plain to see

[@mvmlego1212]

@@Im-mv6bf -- No, it isn't. There hasn't been enough driving from self-driving cars, let alone Tesla, specifically, in order for any differences in the fatalities-per-mile rate to be statistically significant.

[@raze867]

Amazing video , learned a lot, hope for more.

[@Dashboarrd]

What a great video. This really showed both sides of driverless cars. While making sure we know the problems that can arise, it keeps us hopeful that they will continue to improve and become better. Hopefully much better than us!

[@salmansalahuddin6305]

The four way stop is only going to be a problem as long as humans drive cars but when all cars are self driving it wouldn't be a problem.

[@lukealdrich]

Didn't expect to watch the whole video.

[@agustinvenegas5238]

I told myself "oh, interesting, I'll click it and watch it in the morning" The video just finished playing as I'm typing this. ..

[@ramonpinyataro8255]

Great video and information provided -Look forward in getting my self driving vehicle in the future

[@ptsg]

It has been stuck in my mind. 2100%. 21 times faster processing. Holy moly that’s insane....

[@memau]

Quiet disappointed that the focus of this video is so strong on tesla. I understand they're amazing at building the hype but there are so many companies going for self driving tech

[@grenzviel4480]

the focus is "who has the biggest advantage at the moment", which just happens to be Tesla, which he elaborated

[@superdau]

Tesla has the most advanced and by far greatest number self driving cars on the road. All the others are more or less in development and thereby it's hard to tell what they actually do. Also most of the video was very general about what self driving implies, but using Tesla images because they are pretty much the only thing available.

[@TheMagicJIZZ]

They are the only automaker ahead.

[@nimistar01]

until the human element is removed from the equation you're not going to get safe self driving cars.

[@joesterling4299]

If a human is in the equation, the result can't ever be Self-Driving Car. It's something else. And if "something else" means pay attention as if you were driving, when you're not really driving, then the result is way more unsafe than a regular old-fashioned car. Because people don't stay nearly as alert when they aren't actively participating. That's just human nature.

[@paulbyatt3219]

@@yozul1 Spot on. There are millions and millions of driving enthusiasts that love driving. Driverless cars are a way that governments are going to control humans. It's just the beginning.

[@mojave7604]

This is an excellent video with very clear and accurate information.

[@fraserhenderson7839]

Highly informative and entertaining! And an excellent segue to the promotional content.

[@deep.space.12]

2:34 so the cyclist kept walking... headlight of the car was working, right? two programs failed that day...

[@JHA854]

100% the cyclists fault. If someone had been driving, they probably would have hit them still.

[@ivandiaz5791]

@@JHA854 The pedestrian (she was walking her bike, not riding) was in an unmarked crosswalk. She legally had right of way.

[@ivandiaz5791]

She had every right to keep walking, Arizona law like most states in the US states that every intersection of roadways or sidewalks creates a legal crosswalk where drivers must yield to pedestrians, regardless of visible markings.

[@justin60222]

I didn't know about the 4-way stop thing lol

[@feynstein1004]

Me neither

[@iwolfman37]

Bruh... These things are on the driving test you take to get your license. Did no one study for this but me ?

[@justin60222]

Nah it wasnt and either way nobody follows this rule, everyone just goes when they can somehow

[@skirata3144]

I don't even have 4 way stops where I'm living, it's either that one street has priority or in Areas with a high traffic density you put a traffic light.

[@Willam_J]

Happyface - This, and a lot of other useful information, are often neither taught, nor tested for, in driver’s education classes. For instance, when looking at road signs, which indicate an upcoming exit, the direction of the exit is indicated by where the exit number is located. If you have to exit to the right, the exit number is located on the top right of the sign. If the exit is to the left, it’s located on the top left. I only found out about this, about 25 years ago, while reading an article titled “Driving Rules That Most People Don’t Know About”. Over the years, I have told many people about it. No one has ever said that they already knew this. Some have later told me that they suddenly started noticing this, after I told them about it. That would be some great information to have, when starting out driving. You are right, that ‘four-way stop rules’ (and many other rules) SHOULD be taught in driver’s education, but it often isn’t.

[@illsaveus]

I really appreciate you breaking down the strengths and weaknesses of different sensors. I wish you went a little deeper in that area bc it's fascinating but this was great!~

[@theyphridge]

I'd love to see a video on the engeering problems, and efficiency of boxed wing aircraft. Also maybe the X-33 and SSTOs.

[@Sylfa]

I don't think you need AI drivers to be better than an average human, you need it to be better than the human drivers that are crying due to a breakup, drunk, falling asleep, playing games on their phones, etc etc. The sooner we get these kind of drivers off the roads and into an AI car the safer the roads will be, resulting in fewer accidents. Anyone thinking human drivers are better, please check out the dash cam footage on youtube, for instance this channels: kzbin.info A road with only AI drivers will be able to drop to near 0 fatalities even with the current AI's. Combine that with car to car communication being mandatory and the cars will *know* that any object on the road is either a car or something that can be a hazard. No need to identify foreign objects accurately since it can't be a car following the rules if it doesn't have the appropriate transponder. Only mixed roads, without transponders, will require the AI to be better, and even then you could make having a drivers license harder so you get rid of the unsecure drivers, as well as require transponders in even normal cars. Preferably with added eye tracker for the regular drivers so the cars can alert the AI cars if a driver has a medical issue (heart attack, falling asleep, drunk) as well as automatically forcibly slow down to safer speeds...

[@dhiegov]

> Preferably with added eye tracker for the regular drivers so the cars can alert the AI cars if a driver has a medical issue (heart attack, falling asleep, drunk) Imagine the privacy issues that that would arise.

[@Sylfa]

​@@dhiegov"Imagine the privacy issues that that would arise" Your in a box with windows, fairly sure it counts as "in public", hence why the cops will arrest you if you have sex in a car. It's less the optimal solution and more a stop gap though, driving isn't a right so it's not hard to increase the requirements to ensure safety, having an eye tracker to prevent accidents when a driver falls asleep wouldn't require transmissions, simply AI.

[@dhiegov]

@@Sylfa > Your in a box with windows, fairly sure it counts as "in public", hence why the cops will arrest you if you have sex in a car. Indeed, you're right.

[@OfficialFront]

my cars transponder just broke, what does my car do now? what do other cars do now?

[@Sylfa]

@@OfficialFront It detects the breakage, alerts the person in the car and slows down, replacement car and hauler can automatically be sent to the last known location after a connection over the mobile network, your trip can resume in minutes in a rental while your car gets taken to a mechanic without a real person needing to do anything as the hauler is selfdriving. The other cars still detect non-cars and would determine that it's a car with a fault, emergency lights would likely be enabled as well as that is a separate system and would give the car extra room while it navigates to a safer spot. The cost would be much less for a rental system like this than the loss of income of the yearly deaths in traffic causes, so even if the government took the full cost of the replacement car system it would still be a net positive. If transponders are found to be troublesome due to breaking you can have a redundant B-system that allow you to continue the trip and the car will be able to go to a mechanic at a more convenient time, keep in mind it's self driving so it can do that while your at work or out of town without any problems.

[@Mr6Sinner]

Remember kids, if you’re going to have to brake hard to avoid hitting and killing someone: don’t.

[@frozenthorn9619]

As sad as it sounds, there are many situations where this holds true. The uber incident is not among them though, that was just bad programming.

[@patrick7975]

Great and thorough video, thanks!

[@unculturedswine7993]

Unfortunately the misconception a lot of new owners of self-driving cars have is that they can just stop paying attention and fiddle with their phone or do their makeup, when in reality you absolutely have to still watch the road and most of the time at least keep one hand on the steering wheel. You need to be to take over as soon as possible, because eventually a crash or near crash experience is going to happen(there is no if, it is definitely a when). The technology just isn't advanced enough yet for us to stop being drivers and become fully a passenger.

[@kornbread5359]

I did it!! Threw big rock on gas pedal rip neighbors kid

[@MehNamesKing]

I'm not sure why people are hesitant in letting cars drive themselves. At least a car knows how to drive, unlike most of the human "drivers" lol

[@fearofchicke]

MehNamesKing self driving trains still aren’t a thing and they’re literally on rails.

[@VuizTV]

@@fearofchicke self driving trains are common.... DLR in London is one example.

[@ichi1082]

Because in your own mind, you know how to drive but others don't.

[@MehNamesKing]

@@ichi1082 That is fair lol Wouldn't self driving cars be all on the same page? They'd have pretty similar programming to each other.

[@OKANGUVEN99]

I think most people are okay with other cars driving themselves but they want control over their own car. They now that other drivers can be assholes yet they think that themselves are pretty good.

[@voltgod]

Once there are enough of these cars on the road, and they can start talking to each other, it will be a no-brainer. It'll be so much safer, insurance companies will virtually mandate it by making normal cars too costly to insure for most, and so much more efficient getting from point a to b, that buyers will want them. It's not a question of if, but when. Great video. Thanks!

[@euromaestro]

At 3:12 you say the driver did intervene “a second before impact by engaging the steering wheel and braking.” You also mention the very slightly reduced impact speed. The steering wheel was engaged 1 second or less before impact and the brake was engaged 1 second AFTER impact.

[@marloyt7786]

Tesla has: Electric Sedan Electric SUV Electric Sports Car Electric Semi-truck What other vehicle Tesla needs to Electrify?

[@MarloSoBalJr]

Bus

[@TheHua89]

Motorcycle

[@General12th]

Dildo

[@josephb.3841]

Plane

[@duncanevenson4621]

Road quality is going to be the biggest problem

[@taher2437]

Brilliantly explained video, Keep going

[@leadeeeeer]

amazing video! summarized all vision-based systems in the self-driving vehicle

[@Thebreakdownshow1]

in my humble opnion, it will just be better with the self-driving vehicle once the technology is there.

[@dirtymind9154]

i Agree. IT just need time

[@NoobsDeSroobs]

Dark rainy nights, this system fails. They dont cover eachother in all systems.

[@geohiekim8705]

Ultrasound.

[@2nd3rd1st]

0:57 Selfdriving Cur! Haha, I'll be shouting this insult whenever those Waymo taxis cut me off

[@coquio]

This is so rad. Love your content.

[@laos85]

Self driving is stupid and it will only bring more problems(which is already have). The real solution for safer road is to educate sociality and better trainning tool like the simulator for everyone. Faster computer or stupid sensors will never match a well trained and smart human driver who can adapt to every situation that computer or any advance sensors or whatever could never comprehend

[@TheSupperteen]

And people need to be trained to NEVER get distracted, or be tired, or be bored, or in any way be impaired. Except we can't expect people to be at 100%, 100% of the time. Of course a computer isn't going to beat a fully awake and unimpaired human driver in impaired driving conditions such as when it is raining or the road is damaged, but when on a good road or on the highway the human could become distracted because of boredom. Computers don't get bored, this is where self-driving beats humans.

[@RawbLV]

Teslas have driven over 1 billion miles on autopilot and there were only 3 fatalities. How does that compare to 40 000 deaths from regular drivers every year.... In USA alone...

[@laos85]

@@RawbLV That data doesn't show anything specific. How many sefless cars drove in the us compare to probably millions or more human drivers with different quality and weakness of driving?

[@RawbLV]

@@laos85 Exactly. Humans are so random and prone to error. It's impossible to predict them on the road, especially when it comes to split second decisions that they have to resolve while trying to stay cool and not panic. Also there are way too many people driving with their phones out or just straight under influence.

[@ZachMeador]

nice

[@MonclerAP]

ye

[@nievellk1775]

Your Videos really keep getting better and better

[@kyledammann4284]

Fantastic video. Great explanation