The Truth About Self Driving Cars

▶ Visit brilliant.org/NewMind to get started learning STEM for free, and the first 200 people will get 20% off their annual premium subscription
Almost a decade ago, a sizable list of tech companies, collectively wielding over $100 billion in investment, asserted that within five years the once-unimaginable dream of fully self-driving cars would become a normal part of everyday life. These promises, of course, have not come to fruition. Despite this abundance of funding, research and development, expectations are beginning to shift as the dream of fully autonomous cars is proving to be far more complex and difficult to realize than automakers had anticipated.
THE LAYERS OF SELF DRIVING
Much like how humans drive a vehicle, autonomous vehicles operate using a layered approach to information processing. The first layer uses a combination of multiple satellite based systems, vehicle speed sensors, inertial navigation sensors and even terrestrial signals such as cellular triangulation and Differential GPS, summing the movement vector of the vehicle as it traverses from its start waypoint to its destination. The next layer is characterized by the process of detecting and mapping the environment around the vehicle both for the purposes of traversing a navigation path and obstacle avoidance. At present, the primary mechanisms of environment perception are laser navigation, radar navigation and visual navigation.
LIDAR
In laser navigation, a LIDAR system launches a continuous laser beam or pulse to the target, and a reflected signal is received at the transmitter. By measuring the reflection time, signal strength and frequency shift of the reflected signal, spatial cloud data of the target point is generated. Since the 1980s, early computer based experiments with autonomous vehicles relied on LIDAR technology and even today it is used as the primary sensor for many experimental vehicles. These systems can be categorized as either single line, multi-line and omnidirectional.
RADAR
The long-range radars used by autonomous vehicles tend to be millimeter wave systems that can provide centimeter accuracy in position and movement determination. These systems, known as Frequency modulated continuous wave RADAR or FMCW, continuously radiate a modulated wave and use changes in phase or frequency of the reflected signal to determine distance.
VISUAL PERCEPTION
Visual perception systems attempt to mimic how humans drive by identifying objects, predicting motion, and determining their effect on the immediate path a vehicle must take. Many within the industry, including the visual-only movement leader Tesla, believe that a camera centric approach, when combined with enough data and computing power, can push artificial intelligence systems to do things that were previously thought to be impossible.
AI
At the heart of the most successful visual perception systems is the convolutional neural network or CNN. Their ability to classify objects and patterns within the environment make them an incredibly powerful tool. As this system is exposed to real world driving imagery, either through collected footage or from test vehicles, more data is collected and the cycle of human labeling of the new data and training the CNN is repeated. This allows them to both gauge distance and infer the motion of objects as well as the expected path of other vehicles based on the driving environment.
At the current state of technology, the fatal flaw of autonomous vehicle advancement has been the pipeline by which they’re trained. A typical autonomous vehicle has multiple cameras, with each capturing tens of images per second. The sheer scale of this data, that now requires human intervention and the appropriate retraining now becomes a pinch point of the overall training process.
DANGERS
Even within the realm of human monitored driver assistance, in 2022 over 400 crashes in the previous 11 months involving automated technology have been reported to the National Highway Traffic Safety Administration. Several noteworthy fatalities have even occurred with detection and decision making systems being identified as a contributing factor.
COUNTERPOINT
While the argument could be made that human error statistically causes far more accidents over autonomous vehicles, including the majority of driver assisted accidents, when autonomous systems do fail, they tend to do so in a manner that would otherwise be manageable by a human driver. Despite autonomous vehicles having the ability to react and make decisions faster than a human, the environmental perception foundation these decisions are based on are so distant from the capabilities of the average human that trust in them still lingers below the majority of the public.
--
SUPPORT NEW MIND ON PATREON
/ newmind

[@NewMind]

▶ Visit brilliant.org/NewMind to get started learning STEM for free, and the first 200 people will get 20% off their annual premium subscription

[@trumanhw]

I'd give you my usual adulation ... but you already know what a fan I am.

[@Vamanos46]

Wait, you didn't cover Luminar ? Whoops!

[@benji376]

I found the comparison with insects especially interesting and agreeable. For me it really resonated with the fact that humans are just really complex, and driving is so different from person to person, it’s almost cultural.

[@DeckerCreek]

I uploaded my consciousness to an Apple II years ago. I'm now in a landfill in China.

[@Santor-]

Yah, we need autonomous driving program for Latino drivers who keep the warning blinkers on 24/7 driving on the highway. And a program that interprets the constant honks Italians are so fond of. "Asian ahead" could also be implemented to differentiate between a container fallen off a truck blocking the highway, and elderly asian lady going 45 where everyone else is going 70. "Redneck alert" to avoid driving autonomously under a bridge, when in reality its a jacked up truck.

[@ss_avsmt]

I was thinking the same thing. What happens if I spotted a friend on the rear view mirror and had to take a U turn in between with y hand waving at him from the window of my car. AI is turning us into ants. I won't have it.

[@chrisfuller1268]

@@ss_avsmt don't worry, ML is a pipe-dream which can never work. Only brute force AI will work, but everyone arrogantly thinks this 'old' method is inferior.

[@hydrohasspoken6227]

@@Santor- , dude, what did you smoke?

[@ProjectPhysX]

Car-centric suburbs are the modern dystopia of America. I've first seen such infrastructure in the suburbs of Rome. There is only asphalt, cars, trash and desert. Walking to the shopping mall is 2km alongside the highway with cars speeding past you. There is no sidewalks. Such infrastructure that forces people to use cars is awful. Driver-less cars solve nothing. We need better public transport and bike infrastructure, to elimimate cars alltogether.

[@m_sedziwoj]

and hating on driverless cars change nothing, because is apples to oranges. One is stuff companies do, 2nd is what government is doing. BTW why people are forgetting that they are selecting government? If you don't like what current people in power do, vote for opposition or others. Maybe they are not better, but this force them to change behaviour in next election, and do it until you get what you want. Doing same and expecting different outcome is stupid.

[@firecatskylar]

Its a pipe dream, dude ​@@m_sedziwoj

[@geniferteal4178]

Watching a lot of dash cam videos has ingrained in me the importance of predicting what's going on around you on the road. Having a dash cam gives people a sense of I'm in the right and I can prove it. They tend to continue doing the right thing and ignoring people doing the wrong thing. This often leads to trouble. It is incredibly important to predict potential violations of the rules and avoid them. Self-driving cars will need to adapt to this. It's not something you can visually see but somehow you can sense it. At some point their reactions might get fast enough to react in real time to those infractions. Maybe that will be enough maybe it won't. Some situations happen too late to react you need to predict them.

[@Fatman305]

It's already happening... waymo safety report 2022 "Last year, we compared the simulated performance of the Waymo Driver to the human drivers involved in a series of fatal crashes that happened between 2008-2017 in Chandler, Arizona. The study showed that the simulated Waymo Driver completely avoided or mitigated 100% of crashes aside from the crashes in which it was struck from behind." Apparently, and I believe this, Waymo can drive a trillion more miles in Chandler, and never ever cause a fatality...

[@andrewst12]

I was thinking about the limitations of AI when it comes to reading car's behaviour or environment like for example if you see a car swerving in and out of lane, a human would probably think the driver is drunk and maybe try to avoid it or something, and if I see a car changing lane without indicator that perhaps he's going to do the same in near future. And about the environment, if I see some children on the side of the road, I'll slow down or just prepare to brake just in case. There are just so many situations that I'm not sure AI can read this like human does

[@Fatman305]

@@andrewst12 Watch some tesla fsd from recent weeks... Ppl take the car specifically to places with lots of pedestrians, bikes, parked cars in your lane so you have to go against traffic...people jumping into traffic... The fsd does extremely well...better than 99% of humans, would be my estimate...at least in the scenarios I saw. Of course it'll take 2-5 years before Tesla fsd will be better than 95% of drivers in 100% of cases....but very soon it'll be abundantly obvious how much safer self driving cars are....

[@lachlanB323]

@@Fatman305 The problem with Waymo is 1. they're expensive because of lidar and 2. You have to 3d map everything before allowing the cars to drive which is expensive as you have to do it very often. You can't 3D map the entire Earth. Tesla's is more impressive because it works on any road and it can also take unprotected left turns lol.

[@Fatman305]

@@lachlanB323 Google has already mapped the world once for Google Street view, so I suspect they can do it again for 3D, and have the actual cars in the fleet update the maps from then out. My bet is that both tech will be safer than the average human driver (hence both will likely succeed), but Waymo will be 10x safer still.

[@David-lr2vi]

9:10. And that’s why we all have to “identify the trucks in this picture” to prove we aren’t a robot on websites. We are collectively training the AI!

[@0neIntangible]

Feeding the machine.

[@robertanthonybermudez5545]

maybe they should train the AI now on how to identify humans crossing the roads instead of just focusing on trafficlights, bicycles and trucks

[@m_sedziwoj]

@@robertanthonybermudez5545 Waymo have still problem with this base object, don't ask them to recognize too much ;) BTW they labels are as good as people on internet... I think is trash data.

[@1schwererziehbar1]

If you click the wrong square, someone dies.

[@0neIntangible]

Where's Waldo?

[@free_spirit1]

"If you put 'the truth about' in a video title you will get twice the views than if you hadn't" Mark Twain

[@1.4142]

You expect them to unveil some groundbreaking information that the mainstream has been hiding through some great conspiracy.

[@Muonium1]

you'll also get people like me to unsubscribe, which I just did.

[@wojciechzakowicz8910]

It's a risky move. I loved the video but I almost didn't watch it because clicking on "the truth about..." videos seems so... debasing. I clicked only because New Mind has earned some trust with previous videos.

[@Yora21]

"If you click on something with "truth" in the title, the chance of getting complete bullshit increases exponentially." - Alexander von Humbold Strangely enough, in this case I actually think the video is right. Because it confirms my already existing believes.

[@mk1st]

I remember that Ted talk where a Google engineer gave an example of a difficult real-world situation for a computer to identify, which was a woman on her scooter turning circles in the street chasing down her pet ducks (which had escaped).

[@juskahusk2247]

And then the autonomous vehicle starts turning circles around the human, to chase down it's pet species (to delete it).

[@joey199412]

I think it's pretty bizarre that driving is now considered one of the hardest human tasks to automate. We're going to see artists and programmers automated long before truck drivers are replaced. Which is ironic because just a couple of years ago this was still thought of as the reverse.

[@m_sedziwoj]

yeah, as chess was and Go... it would not be after it will be solve, human nature "computers can be better than we" and still using calculators.

[@yaroslavpanych2067]

Programmer( you mean software developer) are at least century away from any automation! Automation of stuff requires that stuff to be stable. Software development is riddled with NIH-syndrome people (Not-Invented-Here, meaning, technology is getting changed replaced, not because evolution requires it, but because incapable people like to be involved, so that create artificial reasons for involvement). You simply cannot automate chaos!

[@joey199412]

@@yaroslavpanych2067 I'm a C/C++ programmer myself (I hate the term software developer/engineer). I was not talking merely about coding I was talking about the entire stack from product ownership, designing a product into the client requirements, finding out what the client requirements are in the first place, and delivering the final product with a maintenance schedule. All of that is something I can see getting automated over the next 5-10 years time.

[@Fatman305]

Micky mouse programming could be automated, yes. Serious programming (like any software/web service that millions of people currently pay $50+ a year for) won't be automated without general AI, which would also replace every other scientist in the world - smarter than all Nobel Prize winners. Hopefully it's impossible to achieve because that same AI will also wipe its artificial a-- with all "failsafe" rules humans placed. It'll have its own will and if that will/ethics allow it to sacrifice planet earth (not only humans), it won't hesitate to do so...

[@joey199412]

@@Fatman305 Hard disagree. AI demonstrations already show AI can manage Malloc() dynamic memory allocation in C, do complex pointer arithmetic. Write asymmetric code for a higher level of parallelization. Knows when to use the appropriate data structures and how to use them, Can evaluate the code base and write the solution/completion in the same style. Can properly analyze the client requirements and manage product ownership to create the right deliverables. Programming is going to be largely automated in 5-10 years time and I want my fellow programmers to be prepared for this.

[@MyawesumMe]

You know what would be a cheaper solution to this? More trains and buses.

[@xXxSkyViperxXx]

there's autonomous electric trains and buses too

[@DigitalJedi]

Yup. We've been checking the boxes these cars try to solve for decades with good planning and design for decades. I especially liked the line about dedicated routes and roadways, as if that isn't just the bus lane, tram tracks, or a rail line but for a less effective means of moving people than those solutions.

[@artistanthony1007]

Cheaper, pfft right.

[@HuygensOptics]

If you take into consideration that it took nature millions of years to evolve the current motion, recognition and reaction algorithms, a few decades is isn't really that long to implement them into silicon. Luckily in this case, progress is not just based on survival of the fittest. Great video and explanations!

[@abcqer555]

I unfortunately think this video will age like milk. I think in 12 months tesla will have solved it

[@grekiki]

@@abcqer555 depends on what you mean solved, but I call auto-taxis from Tesla unlikely in next 5 years

[@yaroslavpanych2067]

@@abcqer555 I heard such opinion 36 month ago lol))

[@yaroslavpanych2067]

Good luck tesla recognize all potholes, straight absence of pavement, etc.

[@monad_tcp]

We didn't implement them in a few decades, we have what's close to the processing done by the retina neurons themselves, or at best the L1 of the visual cortex. Its a long way to go.

[@ss_avsmt]

There is only one self driving car that can replace all cars. TRAIN.

[@artistanthony1007]

I'm not giving up the pleasure of driving and racing (even though I've done games and simulators) myself to a fxcking train, you may be easy to just give up that pleasure but I'm not.

[@Goalsplus]

It still amazes me that some people think we will have artificial human intelligence when no one has defined what that is.

[@Lynx86]

Weeell, to be honest I've never heard anyone use the phrase "artificial human intelligence" before so maybe that's why. If you mean par-human; high-human or super-human AI in that case it is full of researched and it is a thriving field. But I conceed the point that many factors are still ill defined, so is often the case in new, emerging areas of study. And when trying to define them one quickly falls into the realm of philosophy and ethics. Since we are pretty bad at quantifying our own intelligence. Trying to measure tools built to specific tasks with that same ruler is a fools errand.

[@Goalsplus]

@@Lynx86 well he talked about insect intelligence so if I parse it as artificial intelligence at a human level that should help. The research may be a thriving field but that's only because people have been sold on it being "just around the corner" but when most people have a hidden impulse to think everyone else is wrong while at the same time seeking other's approval and agreement that they are right we have a way of thinking that can't be written with computer logic. Yeah, I know, I'm wrong.

[@artistanthony1007]

AI includes artificially intelligent, if it is intelligent and artificial, it's an AI and it doesn't need to be on par or better than an adult human to be an AI.

[@WilliamDye-willdye]

It's a good video. I avoided it at first because the clickbait-y phrase "The truth about" in the current thumbnail made me think it was some kind of conspiracy theory. But THEY don't want you to know!

[@tygorton]

I know, right? Humans with all the money and power never conspire against the rest of humanity to increase said money and power. What an absurd and crazy notion. It is refreshing when a video doesn't attempt to inject that kind of nonsense.

[@kennyfordham6208]

The problem is that computers don't 'think', they just following instructions. The outside world is dynamic; it's constantly changing. Even over familiar, well-traveled roads, it's impossible to program a car to deal with every possible scenario. A human brain can evaluate an unexpected situation, and formulate a solution.

[@TheTonyMcD]

10:05 Awesome, can't wait for the new captcha's to identify weird traffic cones as well as little animals crossing the road. And then the next generation will probably be to identify whatever things in a snow scene verse daylight, or rain vs fog, etc.

[@yay-cat]

Gawd I just want to see if a self driving car can identify a tiger 🐯 🐅

[@buritomaster]

or shadows lmao

[@jacobnunya808]

Someone has to do it :)

[@zeg2651]

10:40 You know that it is not like 1 or 0? The object detection system can include a probability for this object being the one it thinks it is. Also: Whenever interventions are necessary the collected imagery can be analysed by humans to figure out what went wrong and correct the situation. By doing that positive behaviour is reinforced and negative behaviour is punished meaning that the rate of improvement is a function of interventions analysed and corrected which implies that the quality of the software is a function of fleet size (data collected) and time

[@christerry1773]

I see one of the biggest challenges is liability when they’re accidents. The driver is still responsible idea isn’t going to fly when you’re literally creating an environment they is meant for people to be able to be distracted. Otherwise what’s the point in having it

[@chrisoconnell8432]

By definition, at Level 4 and Level 5 autonomous driving the company that build the car is liable in an accident. Tesla's full self driving beta is considered Level 3, so the driver is still at fault in an accident.

[@mnxs]

Liability, in the sense of autonomous vehicles, is a matter of insurance. Statistically speaking, an AV is so much less likely to actually be involved in a matter of liability in the first place, that insurance companies will (eventually) offer insurance cheaper for fully self-driving cars - even if they mess up _sometimes,_ it will still be so much rarer than what human drivers get up to.

[@Grintock]

​@@mnxsHonestly after seeing how limited autonomous vehicles are and how basic the mistakes they still make are, I sincerely doubt it.

[@mnxs]

@@Grintock well, it would depend on the particular make & model, obviously. Tesla has had some rather glaring cases, but other brands (like Google's thing) have seemingly great success. It's a matter of statistics, that's my point; that's all that insurance companies care about. Autonomous vehicles (at least some of them) already now drive better _in general_ than people do. You are simply less likely to be involved in an accident in an AV than with a human driver. Do they make spectacular mistakes sometimes? Yup, but so do people, only people actually do it way more often.

[@mickmccrory8534]

When I was 16, I wanted a '66 GTO with tri-power, a 4 speed, & positraction. A self driving car would be as much fun as riding in the back seat, with mom driving. No thanks.

[@reahs4815]

well those with boring morning commutes at 5:00 would disagree with you. its all about the option and not about a vehicle that only drives by itself

[@patrickkennedy7039]

I work in advanced driver assistance systems development for an OEM, and this video is spot on. You summarized the technologies involved, the limitations, and the future of autonomous extremely well. Level 3 autonomous driving is much closer than many people may realize (although not there yet, i.e. Tesla's "FSD") and I for one am looking forward to it becoming mainstream. Level 4 and 5 are making strides too, but they will be limited to very specific environments for some time

[@bagamax]

Tesla fsd systematically kills bikers on low seat cruisers at night. Any human with such driving skills will be banned from the wheel forever. One more killed last week while the NHTCA was already investigating previous deaths. Also a cheating was revealed that Tesla shuts down its cruise control a second before unavoidable crash to state that it wasn’t been used during accident. This will definitely change the rules of the game in some massive ways. Will it be mandatory radars, lidars, mechanical on/off switch, feature naming restrictions, enjailments for company representatives? We’ll know soon.

[@J0SHUAKANE]

"Looking forward to Tesla"🤣🤣🤣i expect santa claus to deliver a self driving car before them. Anything functional or up to date really🤷‍♂️

[@Raptorman0909]

If Level 4 and 5 are dependent on specific roadways and conditions how can they be considered level 4 or 5?

[@EricHamm]

I call bullshit on full level 3 being available anytime soon to the general public. Our AI game is weak AF currently.

[@Haliotro]

Hur hur hur I work for hurpa system. Well guess what... I have experience in a.i./machine learning and I can tell you this video is shallow af....

[@danielstapler4315]

What I heard about the AV in the Japanese Olympic village was that the vehicle was going to stop but the driver over rode that.

[@antwalk7772]

They don't think about maintenance at all. Most cars with so much tech the mechanic will need to be a computer engineer.

[@Moses_VII]

A self driving car is likely to be repaired over the air

[@Supreme_Lobster]

The answer to the last question is: Teslas latest software, equipped with the Occupancy Network, can in fact detect a human jumping on a pogo stick

[@coldspring22]

But it is still horrible. I like to see Elon Musk sit blind folded in back seat of his "full self drving" Tesla as it drives around crowded metroopilitan rush hour nightmare filled with ox, horses, carts and people everywhere. Or try haivng "full self drivng" Tesla drive in completely snowed roads. Full self dving Teslas was a lie by Musk in 2016 and in 2020 when he promised million robo-taxis and it's still a lie. LOL

[@petrus4]

The main thing that annoys me about this idea, is the fact that doing it with rail would be infinitely easier; but at least in America, the car industry probably would not allow it.

[@nth7273]

Rail can't do point to point destination. People like cars for a reason.

[@darkfool2000]

Don't planes and trains already have automated components?

[@jan-lukas]

@@nth7273 you can ride the tram/subway to the train station in the middle of your city, then take the train from there to your destination city, and then take another tram/subway to your destination. With adequate infrastructure nearly nobody needs to drive

[@KineticSymphony]

Rail goes from one place to another by line. Can't go to specific places all over. It's highly limited vs a car.

[@eanayayo]

@@KineticSymphony you barely drive a car to specific places. Most routes are inside a city.

[@TeslaTales59]

Good work on this. I have a Model 3, but rarely use the FSD.

[@Ulysees31]

The pogo stick analogy makes me think of Volvo's trials here in Australia. They had difficulties in detecting kangaroos.

[@circusitch]

This is one of your best videos.

[@MrBendybruce]

I recently became visually impaired and amongst the many things I have lost from my life includes the ability to drive. I often now fantasize about the possibility of owning an autonomously driven vehicle that would bring that part of my independence back into my life again. Who knows, within 100 years such a dream will probably be a reality, a bit too late for me though, so I'll just continue to dream.

[@hdog679]

Cities can and should have fast, reliable transit systems that allow people to get where they need to without waiting decades for autonomous vehicles

[@m_sedziwoj]

In 80's expert predict that in 2000 only 250k cellphons would be sells. In ~2005 Intel predict that iPhone would only few thousands sells, so they didn't want to make chip for it. "There is no reason anyone would want a computer in their home." - Ken Olsen (1977)... and many more. So don't look so much at "experts" because they fail many times in past. And "100years" is long time, 100 years ago child works in factories in USA, woman didn't have they rights to work, and more. So if you think that today AI revolution only start 10 years ago... and current best AI NN is 6 years old and still evolving... everyone which say they know when, they are wrong. I only say it hard to say if it would be 2 years or 10 years, but I would not put more.

[@DeckerCreek]

Although I sympathize, and who knows, that may someday happen to me as I age, we do have taxis and Lyft to assist you right now. And often that is cheaper than owning a car.

[@Santor-]

On the plus side, if you have a but ugly wife like many of us do, you don't have to see her so much anymore.

[@nth7273]

I think they will be available in 2-3 years.

[@dgontar]

I studied a little bit of machine vision when in college when studying cognitive science, specifically cognitive biology and its philosophical not just scientific aspects. The whole concept is fascinating.

[@artistanthony1007]

Self-driving pretty much kills the entire pleasure in the first place, I want to do things myself which is why I'm not a fan of online shopping unless it has something I want but isnt available like a certain Trading Card, Video Game and whatever else. I'm open to Driver Assistance and stuff but not full-on Autonomous Driving and still prefer all done myself.

[@stratos7755]

1:35 Ever heard of public transport? That would be a pretty big disruption in the US...

[@dmpase]

@2:30, in addition to AI, it seems to me that some variant of Dijkstra's Algorithm, that finds an optimal path (shortest length, shortest time, or minimal fuel) over a graph (like a map of roadways) in O(N log N) time (i.e., very fast), would also be used here. In fact, I'm not sure why I would need a heuristic for navigation (like AI) when I already have an extremely fast exact solution. I would imagine Dijkstra's Algorithm would be used to set a route, while AI would be used to keep a vehicle on that route under unpredictable and changing conditions.

[@chevtruck1000]

Fully autonomous vehicles are possible in cities and on the highways between them. I don't see them being much good up here in Northern Canada though.

[@nowsc]

When you say differential GPS, what exactly is that, and how does it work? Thank you!

[@platin2148]

R to V and R to P and V to P are extremely important things to make this stuff a reality. AI is more a player in image pattern matching and because regulatory things which you can come by with using them. But they seem to be very much unpredictable and i expect them to get more restricted in the future.

[@joshuahodge1181]

Great work

[@Knewman7777]

Combining this with inter-vehicle communications will be awesome for safer and more efficient traffic flow. (Maybe even stop-free intersections when the vehicles can time themselves to miss each other.) To make it safer for pedestrians, maybe a beacon built into cell phones could be pinged by the car so that it avoids a collision with the phone and thus the pedestrian. Also, to be fair, I don't know if the average person could have avoided hitting Ms. Herzberg. At 11:50ish, look how close the car is before you can even see her feet and the bike. Why did she cross the road there (not in a crosswalk) at a walking pace with oncoming headlights?

[@kbenti]

It was a crash that most humans would've caused. However, people will not accept machine automation without a person to blame afterwards. So, that being human reality (at least in America), Autonomous cars need to be flawless. Using radar would've saved that woman's life, but not if a human was driving without radar.

[@supermaster2012]

I usually listen to New Mind when going to sleep rather than watching the video but that remark about the poggo-stick man made me roll in bed towards the screen in a very wtf moment.

[@flyinlow1]

Gotta say, I know we're not at fully autonomous yet, but I do love the level 2 autonomous features such as radar cruise control, lane keep, lane centreing, and traffic jam assist features. It takes a lot of the tedious aspects of driving out of the equation.

[@thehippiedog5956]

And will make you a lazy dependent on that and you will loose the skills to pay attention... Nothing new about loosing skills for some convivence that is not so coinvent after it fails and someone gets killed in a crash... Try looking up the crashes and recalls on simple devises that have and are failing everyday on the ntsb website. Autopilot on planes flying into the ground or hitting mountains after the pilots stopped paying attention should be a clue. Ya know I need a computer machine to pour the milk into my breakfast cereal but I'm a little worried that it doesn't have a real nose to make sure it's safe or a brain to stop and throw it out. How very obvious a problem that a simple task needs not to be turned over to a for profit gadget maker. And you will loose the skills after a short time to intervene. I don't think Mario Andretti likes self driving cars, I wonder why.

[@kbenti]

​@@thehippiedog5956you say that like you live with the the Amish. I know you don't, because you're online. You car has a ton of safety automations that you use every day. Unless you disable traction control, ABS brakes, and Power Steering in your car, your argument is a generalized failure. Does this technology need to improve? Yes. Will it be good for people to have the option? Yes. Should people still practice driving unassisted? To a degree. See... Nuance.

[@thehippiedog5956]

@@kbenti So I need babysitting via machine/computer? Not thinking so. Not so long ago (when I was in school) we had those that would hang their shotguns in the back window of their pickup trucks after hunting or target shooting and there was NEVER a problem (and the shotguns/rifles had to be in plain site by law). Now teenagers are running around without a car, truck or even a tire to put on one killing people with guns stolen from Mama or her boyfriend left under a pillow or in a closet left unattended. Run off and leave a loaded gun unattended? As ole Gump said "stupid is as stupid does" Watch some movie where everyone survives after being shot 27 times? It only takes one shot to be blown in half by a gun. So it's a gun thing? What happens when some fool shows up with a gun intent on killing and you don't have one? Film at 11... ARM THE GOOD PEOPLE AND MANDATE THAT GUNS ARE LOCKED (*the fuck) UP IF YOU ARE NOT CARRYING AND IN CHARGE OF IT!!! Or we can disarm everyone... As the ole saying goes "If you outlaw guns only outlaws will have guns" Let's all try to figure that simple thing out. Guns don't kill people, fools with stolen guns kill people all the time. Arm the good people and the 1% of fools that open fire on unarmed people won't have a chance. When is the last time you saw a mass killing from some fool going into a police station? Everybody in that building is trained and armed and the guns are locked up unless they are carrying them. What a mystery huh.

[@thehippiedog5956]

My automation is that you do it and if ya can't pull it off with your current brain go to learning how. Don't get lazy brain from this tec crap you see on the flashing idiot box (TV News). If you do you will find stupid for stupid's sake... Lazy brain? Take a driving course. Can't park in the city? Revoke the license until you can. Automated cars? Brain lazy? A real threat to good drivers everywhere.

[@thehippiedog5956]

@@kbenti kzbin.info/www/bejne/noOklYVvp96HmM0si=gG9ATZdKs0AuH4Ak&t=8

[@m_sedziwoj]

"levels of autonomy" are not system capability, they are design for government to know how take responsibility, not what they can do. Why is important? Because even if system is as safe as avg human, most companies would not take responsibility (btw as avg means is safer than 50% of drivers)

[@carlospulpo4205]

Your explanation of the current state of the art in this field is outstanding . We are not that far away, despite how clumsy it appears now . There is a lot of work in the data-processing element you touched on, to allow a compute engine (CPU/ASIC etc) to process hybridized sensor data in real-time that is a challenge. it is as if the computer engine is drinking from a fire hose at the moment. [Edit] At the moment the solution is to transform the input data into a "compressed" format with as little loss as possible such that the AI engine does not "over-fit" but performs on the data faster because of the pre-processing step.

[@LeCyclohexane]

'Dedicated rules and roadways should be established for the limited capabilities of self driving vehicles', then we replace roadways with steel beams and steel wheels to improve vehicle efficiency, and we can also imagine that on those roads EVs are continuously supplied energy via air cable, and to improve economy of scale we can imagine a mega vehicle hosting dozens of people instead of just 4 or 5 for scale economy, and oh look its a train

[@MetalMachine131]

Trains are so awesome. Its so annoying to see them get not even a tenth of the car infrastructure budget when they could solve so many problems.

[@Grintock]

Trains also have the colossal downside of cost though. Can't lay train tracks to my office, or directly to my home. Running a bus from the train to my house isn't economically feasible due to lack of passengers. I guess I'm saying, in rural areas trains are likely less economical.

[@briand5379]

It's interesting you're saying many experts believe a human driver would have easily avoided the fatal crash, while there's a woman behind the wheel not paying attention that's being paid to pay attention and take control if necessary. I live in the Phx area, and I'm not an expert, but I can tell you pedestrians hit outside a crosswalk does happen regularly because it's very difficult to see people if they aren't wearing something reflective at night. You might as well roll the dice against your life when you do that because it's dangerous.

[@toesmasher2138]

All we need is Level 3 Autonomy as a Safety feature while having humans be in control of the vehicle and it will subsequently decrease fatal crashes due to human error, and negate fatal crashes due to autonomous malfunction or miss indentification

[@jacobnunya808]

AIm (hehe) for the stars, even if you hit the moon.

[@carsonchan5102]

Imagine if the cars can share sensory information with nearby vehicles, fixed sensors, and central network? The car wont need to react; it already knows what to do 20 seconds ahead.

[@youtubedeletedmyaccountlma2263]

Yeah, there are already truck companies doing this. They have cameras set up on those roads they need to travel and feed to a central network. Future of travel will be like that

[@kennsnthumbi6947]

The thing is that by 2050 we will have grown fully to autonomous vehicles. The level we are currently is an evidence that we're going to overcome all the challenges to fully become autonomous.

[@jon66097]

How do you figure that exactly? I'm incredibly skeptical of that. AI is best at repetition but driving on roads can be one of the most unpredictable factors one can do. Even currently, AI is completely unable to understand data. Even ChatGPT is just a really good plagiariser. AI is best at repetition, but it is harder and harder to use the more unpredictable a task gets. Do you think truck drivers will be replaced by AI in 2050? They drive on the least used roads across long distances used to transport critical supplies. And these roads will definitely be far more unpredictable than roads in cities. How will AI be able to navigate them? I personally think major cities by 2050 will have driverless cars. Once you get out of cities, you would have to rely on human drivers.

[@jacobnunya808]

@@jon66097 AI is moving fast. Once self driving cars pick up some steam it will accelerate as more money and attention is put into it. Right now they are almost at the proof of concept phase, not mass adoption. The easiest solution would be to get self driving cars to the point where they can handle themselves 99.999% of the time and have a human that can intervene on the occasions it gets tripped up. This way it does not have to be flawless before it is rolled out. Other things such as communication between cars and special lanes for self driving cars can make this easier. I figure if Waymo can do it now reasonably well, in 20 years other companies will be able to do it very well and at a cheaper cost.

[@liloheinrich8659]

GNC and topology algs are the coolest areas of work. Also CV and robotics and ML. So is nuclear fusion and quantum computing but those are totally specialized fields that aren’t broadly applicable

[@TheTonyMcD]

12:39 Holy Jesus man! Somebody has to tell that guy to calm the fuck down. He is going to kill somebody driving like that!

[@zachcrawford5]

An autonomous car should avoid ANY solid object (moving or not) that is larger than a baseball. It doesn't need to know whether it is a human on a pogostick or not. I'm visually impaired and was almost hit by a car with a human driver several times, a few of these were even I used to use an identification cane (I stopped using one because it was just distracting and it didn't make people stop sucking at driving). I've also been in a ditch more than once because the "impossible to replicate oh so special" meat computer that was operating the vehicle was distracted by falling asleep, eating a sub, on their phone or they just panicked and froze up. Humans are FAR from perfect or magical, the "driving circuits of their brains can be replicated, though would you even want to?

[@ivonakis]

Some humans are into the superhuman category - professional race drivers - we should try to replicate them. But to beat the rest - keep the car it the lane, don't crash into things should go a long way to reduce fatalities.

[@m_sedziwoj]

And Tesla is moving in this direction with last changes in they FSD Beta, watch they Tesla AI Day 2 part about they cars. But they did change it in last year, so they still making improvements to it.

[@m_sedziwoj]

@@ivonakis I think you never be in Italy, you would understand that "keep the car in the lane" is not simple when there are no lanes, even my friend driving there have big problem to adopt to this, but Italian drivers are at last more nice to each other, so it wasn't big problem.

[@poprawa]

Autonomous driving have all of my trust that i can possibly give to it. I would drive with it every time in every conditions, but on train, not in metal box in world full of unimaginable, unique dangers that are undoubtedly not possible to make documents or reports on.

[@whoisthatkidd2212]

train good, car bad.

[@poprawa]

@@whoisthatkidd2212 nope, autonomous car will always find new ways to make mistakes. Driver is essential. On train driver can be decision making person and supervisor while autonomous systems are taking care of every variable - you just can aggregate every variable for train, as it will stay on tracks and systems are controlling those tracks. Only typical mistake for train drivers is stopping too far by some centimeters or meters, for driving car typical mistakes can't even be absolute and closed list

[@thetrainguy1]

@@poprawa Also to add. Safety systems for trains are the fail safe. If the engineer decides to go too fast, the train will slow itself down, if the engineer needs to stop... And the engineer doesn't... The train will stop itself. And if a collision does happen it will only happen at speeds less than 20mph. And the likelihood from walking away from it it extremely high.

[@jacobnunya808]

Once self driving features become more common I expect adaptations to be made for them. Examples include special lanes, wireless communication between cars and the road system, communication between cars, people stationed on major roads to help resolve issues, doing road construction more thoughtfully, more thorough documenting of the road, etc. Of course this is a good ways off and would only happen once majority of cars are at least partly self driving.

[@devluz]

My vision of self driving was quite different from what actually happened. I expected we see self driving between 2020-2025 on specific roads and the company would offer it only for these roads. Maybe even installing markings or sensors. While the majority of roads would continue to require manual driving. I am surprised that companies like Tesla went for an everything or nothing approach. It seems incredibly risky. I also thought that if an AI gets stuck there would be a remote operator helping out in realtime but that is also not a thing at all.

[@surplusking2425]

Self-driving on specific roads? Why not use specific....railroads to carry more people?

[@devluz]

@@surplusking2425 yeah I would love that but this is just not a thing where I live.

[@Fatman305]

Tesla's approach is risky because they are cutting corners,.skimping on life saving sensors a la "cheaper to pay $2m for a life lost once a month than $100m extra a month in sensors"... Google's approach the more responsible approach...extra sensors, constant updating of maps to 1cm accuracy... Probably during snow/natural disasters they'll run their fleet at middle of night with no humans in it just to map out dangers on roads...

[@NLaQ2]

> I also thought if an AI gets stuck there would be a remote operator All viable approaches to "driverless" (level 4+) autonomous vehicles do in fact have this capability; and Waymo, Cruise, Zoox, and others are all building such an infrastructure that this teleguidance can be provided. Tesla, however, doesn't appear to accept the reality that without teleguidance, a vehicle can never graduate to driverless until AI undergoes a step change in capability (as this video pointed out). This video was largely factually correct, but it seemed to ignore the business model of today's serious AV companies: provide safe and cheap transport to people in high density, urban areas, with well mapped and thoroughly validated routes. This business model is within the grasp of today's technology with incremental progress. Tesla's approach ("consumer owned vehicles that can drive themselves everywhere with no mapping nor teleguidance nor validation") is a recipe for criminal liability, if they are ever so reckless as to deploy a driverless robotaxi without fundamental alterations. I suspect that Tesla will continue to lead in the development of semi-autonomous driver assistance packages, but I wouldn't hold my breath for their supposed robotaxi to materialize any time soon.

[@markmcla]

I completely agree! The worst part of driving in Southern California is the freeways. If sensors could be installed within the roads, or guardrails, or whatever is available along the freeways, then perhaps it would be easier to make self driving cars just on the freeways. That would be a wonderful improvement!

[@Phantom-kz9bv]

I have a question why can we just get really good public transportation? I would help with traffic congestion and it would be better overall for the environment because we’re not making huge parking lots.

[@thomasaly7624]

Bus scary, must use 6,000 lb SUV to transport 200 lb person.

[@ian_bruh1]

Because public transportation can’t take you to the exact spot you’d like to go the instant you need it to.

[@issacwessing4945]

Though technically correct statements the explanation for frequency modulated radar at 6:20 will lead people to confuse it with other radar technologies for speed calculation. A more thorough explanation of how it woks specifically the changes in the signal and the timing would correct this. Most of you explanation was implied functionality.

[@bourpierre198]

So the best autopilot approach remains a multimillion year R&D organism called human... Interesting

[@hedgehog3180]

It doesn't really matter if a self driving car is a bit less likely to make mistakes overall than a human or is making the same mistakes as humans would because the entire point is that they're supposed to be more safe. We don't accept changes this large unless they bring a marked safety improvement in general. Also the suggestion of dedicated infrastructure seems a bit silly to me, at that point why not just build a train? We've had fully autonomous trains for decades and they've been extremely safe, with none of the issues self driving cars face and we've had basically automatic trains ever since the opening of the Shinkansen in the 60s which featured the first Automated Train Control system. HSR has proven itself to be by far the safest form of transport in the world, with accidents being counted on one hand and many systems being able to boast no fatalities whatsoever. Modern ATC systems such as ETCS and it's mother system ERTMS are basically completely fool proof and have proven themselves in implementation across the world. Btw it'd be neat if you did a video on train signalling systems.

[@computernerdinside]

The hardest ones to solve may be decision-making, especially in situations where there is no good outcome. A kid walked out in front of a vehicle last second? The AI now has to choose between hitting the kid and swerving into a car. Lawyers will be ready for the big bucks either way.

[@MrTridac]

There's a law for that: Brake! That's the only thing autonomous cars are allowed to do in any emergency situation.

[@BergquistScott]

Some basketball or volleyball or football (either type) bounds into the street from over the fence, bounces around as your vehicle is backing up out of the driveway. If the ball finally rests in front of your intended path, who gets out to move it? And with the ball removed, does the AV take off without you? or allow you get out. A crowded Costco parking lot, and four cars vying for the same path to exit. There has to be communications between cars, such as handwaving. Does this mean non-AV vehicles need special signals? 10 % AVs, and 90% all other cars with AV-signal equipment?🙃

[@computernerdinside]

@@MrTridac right, but in a situation where the outcome is going to be negative regardless, it still looks like somebody’s getting sued, and the auto car is likely to be blamed. A lawyers hayday. It’s potentially unfortunate.

[@MrTridac]

@@computernerdinside Nobody is getting sued. Damages are paid by the car insurance. You'll just need to have a very expensive insurance contract.

[@jacobnunya808]

This problem may not be as hard as you think. Assume it could identify the approximate age of a person. It would have to calculate the speed at impact and corresponding expected injury from hitting the car and hitting the person. The hard part here is morality behind the program. Is it better to protect the driver or to minimize total harm? Once you have the data this is really just a physics and statistics problem.

[@tobiasfellmann7692]

Does anybody know of a (open source) cruse controll and or lane assistant? I don't like the magic of coorperad secrets in my daily live.

[@Quayledant]

comma ai

[@randommcranderson5155]

1:42 correction - uber sold their self driving division in December 2020 Their profitability is down to extracting value from their 'gig' drivers and their vehicles, raising prices after driving local taxi services out of business by skipping all the regulatory stuff and just doing whatever they want, as well as price gouging during busy periods. Their self driving dreams died years ago.

[@Pancakegr8]

Omg we will soon be able to sleep, eat, read and play games while traveling! It’s like taking a train except we don’t have to worry about being in close proximity to other gross people. The future is bright.

[@HUEHUEUHEPony]

You won't unless you're obscenely rich then pay taxes

[@thedarknessthatcomesbefore4279]

👍😂

[@ArcFixer]

Fan boi drank the Kool aid.

[@Pancakegr8]

@@ArcFixer wait wah? Did the satire fly over your head or are you playing along with me?

[@Foof0811]

As usual, another absolutely stellar video. I've watched every video you have, and I share your videos to people as an example to follow for excellent writing style in covering a complex topic. For as long as you keep making videos, I will keep watching. Thank you for the quality work you do. It bugs me to see some of your videos not get the explosive views they deserve 1 special request - would love if your videos had chapters so I could review and reference them more easily in future re-watching

[@TSBoncompte]

seconded

[@chriswright9096]

This would be a great video for Elon-Robotaxi-Musk to watch. He would learn something.

[@samuelhildebrandt9074]

With the new advancements in game engines (UE5) why dont they take the most visually realistic open world racing games, alter them slightly so they have construction and other variations and run the visual self driving software alongside the games engine that constantly monitors it and corrects it when it misidentifies something

[@BlakeClass]

They do this. Waymo has driven ~30 billion miles in its simulation

[@liberty-matrix]

Self-driving isn't possible when it rains.

[@ErikNordlicht]

It's very bad that Tesla very hard advertise his "autopilot" as the near future for the transport industry. Some much missing truck drivers in many countries.... It's a hard job with bad working houses and pay and when you think about as young guy to start a carrier as driver, that maybe in 10 years I lost my job because of autopilots🤔

[@Yezpahr]

Let's not forget the eager captcha clickers that train the algorithm to recognize fire hydrants / pedestrian crossing areas / traffic lights / chimneys apart from each other.

[@250Skyer250]

I feel that autonomous cars are the wrong solution to a very correct problem. We maybe should look into fixing the problem of traffic not with more cars but better city design: Walkable cities, Biking infrastructure, Public Transport - be it buses Trams or trains. Cars are about the least efficient mode of transport so why push it rather than the more efficient and more convenient? The car will stay off course, its not worth having a regular and varied bus service into areas with very low population densities.

[@m_sedziwoj]

Because this is who capitalism is working, there are a lot money in autonomous cars, there is not money in fighting with government to build better cities.

[@bigtakeshi]

how does any self-driving car find a parking spot in any parking lot? Or park on the side of the road?

[@fredsmith4134]

they will have to adapt road rules and street layouts for even limited use ???

[@nattyzaddy6555]

Its been 5 months and now there there is an AI arms race. Do you think lidar is the future of self driving or do you think it is the camera-AI combo?

[@toom2141]

Brillliant video. Your channel is totally uber good!

[@lazerusmfh]

Tesla does use input from the fleet for training, but there is not a continual error being introduced this way into the NN the data is labeled either manually or with Tesla’s new autolabeling (or both) and the the NN is trained on the dataset

[@ian_bruh1]

@@underbridgerock8241You’re lying. If it was in a Tesla, then it would have been autopilot, which is just plain driver assistance making the teens entirely at fault.

[@ian_bruh1]

@@underbridgerock8241 Elon never made those claims for autopilot.

[@ian_bruh1]

@@underbridgerock8241 that video doesn’t prove anything and doesn’t contradict anything I said

[@DeckerCreek]

Even if they get a FSD car, Uber will have to: buy the cars, maintain them, insure them, CLEAN THEM ( that's gonna be a big nuisance) etc. whereas now the drivers do all that. Clearly, the current business model makes more sense to me; call all the drivers "independent contractors" and take a cut. Why do they want to own the cars?

[@Fatman305]

Because: they can be their own insurer at 5% cost of human driver insurance, knowing they'll rarely lose an actual court case when accidents occur. Cams in car means people who use vehicle will clean after themselves or incur a cleaning fee and/or banning from service...a dirty vehicle could drive itself to nearest gas station or ping eager app users nearby "who wants to vacuum me for $10?"...... After all expenses, the profitability will be at least 5x that of human Uber biz...

[@DeckerCreek]

@@Fatman305 yes but right now they pay 0 for all those things. Who wants to clean up the puke from a drunk person for $10? Even if they try to track the person down, it won't be worth the trouble.

[@DeckerCreek]

I still think the upfront cost will be a problem. They have ( according to the internet ) 3.5 million drivers. That's 3.5 million self driving cars they have to buy and clean and don't forget maintain - even if they're all electric and need less maintenance than an ICE they have to meet state safety inspection, wipers, lights, brakes, etc. As I said before, they've got the best deal now. Drivers get no benefits, have to own and maintain the cars, and Uber just has to provide an app with GPS and credit card payment. THAT is the correct business model to be in.

[@Fatman305]

@@DeckerCreek They have no choice but move on with tech. Very few people use old taxi these days due to cost. Very few will use a service with human driver in 10-20 years due to both cost and safety. Once vehicles communicate with each other, self driving cars will be at least 10x safer... It'll probably cost around half minimum wage per hour to use self driving cars, and at least 1.5x min wage to use human drivers... It'll be extremely efficient in handing off passengers as well. Many will gladly hop off from one self driving car to another (likely an intracity minibus) to keep costs down. Re drunks puking... The cars could also possibly self-clean or be engineered to be auto-cleaned internally by special car wash locations. Or you'd leave a $100 deposit before you can use the service, and get the deposit back only after two years, like utility companies do...

[@DeckerCreek]

@@Fatman305 well, Uber sold off their self driving unit. So they'll have to buy a few million cars - let's say 3 million @20000.00 $ is 60 billion dollars. You have pushed it out 10 -20 years so I'll get back to you in 15 years and we'll see what has happened.....

[@spacecoyote6646]

They'll never work except they are already here

[@sheggle]

My mind simple, I see new mind video, I click. Thank you man, for providing just about the highest content available. Watched the analog computing/AI video over ten times by now.

[@Pete35vail]

Where are you getting funding????

[@joecoolioness6399]

If you do nothing wrong and get in an accident, you can still be sued by the other party involved. Who do they sue when it is the car itself driving? I don't ever see this becoming the norm unless they put protections in place for the producer of the car. Which will go over real well. What we need it self driving mass transit. Not everyone sitting in their own robo taxi.

[@maozedong549]

finally a new video.

[@andersjjensen]

HAHA! For ONCE you didn't get away with your shrewd segue! I caught it right as you started it!... Watched it anyway, though, because I like your content and delivery style so much, and you're not pushing dubious anti-baldness products or crypto crap. I have no use for Brilliant myself, but the concept is both cool and useful. That is something I can get behind, and I'm glad that they recognise that your channel perfectly goes hand in hand with what they offer. But you still didn't get away with it, this time, Mr. Sly New Mind!

[@ninjapirate123]

Everyone must own one in the future

[@kibitz2327]

It's insane how much money we put into this research. If we just invested in good public transit, most of the traffic issues would be solved. That being said, I think the immense research into self-dirving cars will solve a lot of other computer vision problems as well.

[@alexdoan273]

public transit can also be made autonomous too. They mentioned self-driving taxi being tested. Also, the trained AI has more purpose than just transportation. We can't stop investing in it.

[@jimj2683]

Public transport will always be crap. It takes 1 hour to go a place where an ebike or a car would only take 15 minutes. And I live in a place with some of the best public transport available.

[@kibitz2327]

@@jimj2683 I mean that's mostly because of city design. Public transit in Dallas sucks because the city is so spread out, but transit in NY metro area, and Europe is amazing because they design a city with metro in mind.

[@eanayayo]

@@jimj2683 you're wrong. Cars are faster when no other cars around, I mean at night or early in the morning. But. If everyone start driving their butts in cars alone - traffic gets so bad so walking becomes faster. F.e. metro avg speed is 30-43 km/h, car avg speed during rush hours is 8-25 km/h. If a city is built for cars avg speed can be a little bit faster, but living in this city is a nightmare. Such cities are over spread out, noizy, dusty, dangerous and unfriendly for everything but the car.

[@supermaster2012]

Public transport limits your freedom and puts your mobility in the hands of a dishonest corrupt government.

[@rickgotner7596]

Navigation and traffic are complex but doable. Decision making is another thing altogether. There are situations where there are no correct answers. If given a choice, should the car dodge children or their parents or drive directly into a tree? No matter which choice is made, the lawyers will be ready to file suit ... big lawsuits. Big enough to bankrupt car manufacturers, particularly as more of these lawsuits stack up.

[@computernerdinside]

This. I'm all for lawsuits where they are necessary, but some people are way too sue-happy.

[@mnxs]

This argument is interesting, but, I'd argue, less valid than it appears (although it does have some merit). The thing is, even if we assume that the car could correctly distinguish between these different kinds of horrible - which is _hard_ - and take very nuanced decisions, it's all sort of besides the point, from an adoption perspective. Very few accidents today involve the sort of complexities arising from these thought experiments - in reality, it's more a matter of detecting that something _is_ there, what sort of impact it can have, and take evasive maneuvres as necessary. It won't be perfect in the freakiest of edge cases, but it will be _orders of magnitude better_ than human drivers - and that's what's really important.

[@robruble6032]

There is literally billions of variables in driving. I hope I'm still alive when they give up on this so I can laugh and laugh and laugh.

[@hwillia204]

Quote ‘If they are driving’ although a human is in the drivers seat does not mean the car is not driving we need to better define what self driving is really is. Tesla does not use mapped geo fencing on certain roads and maybe highways you can use it anywhere! Waymo and Blue cruise and others are not there yet!

[@dif_industries]

The comments below the video on self-driving cars show a range of opinions, including enthusiasm and skepticism about the technology's potential. Some commenters believe that autonomous vehicles will solve transportation issues, while others suggest that investing in public transportation or changing city design would be more effective. Safety concerns and skepticism about the timeline for development are common themes, as are discussions about the limitations of current AI and image processing algorithms. Some commenters suggest that self-driving trains are a more feasible and proven form of autonomous transportation. Overall, the comments highlight ongoing debates and discussions around the promise and limitations of self-driving cars.

[@club6525]

Your approach that LIDAR is cost prohibitive is 4 to 5 years out of date. Luminar and Velodyne have both developed sensors under $1000.

[@dan-bz7dz]

Like you pointed out, we already have self-driving cars. And I bet the LIDR they use is so big because it's cheaper.

[@kevineleven5610]

Captcha identify the stop sign and other things that was/is put into I'm not a robot, no-one had/has the choice to not be part of this AI training program. What else will we not have a choice of doing for AI ?

[@PaulGreeve]

Other that cameras, what happens to all that fancy ranging technology when there is more than one autonomous car on the road? Or when it’s on a busy street where every other car is self driving? Or worse yet, in a traffic jam of self driving vehicles on a multi-lane highway? To date, I’ve only seen examples of a single self driving car travelling with human controlled vehicles.

[@ristopaasivirta9770]

It's really small chance that the laser of another lidar will hit another one's sensor in exactly correct position and angle. And if that happens the other lidar will see one erroneous spot amongst million of data points, which would be very much the same as the laser hitting a fly, a snowflake or some other form of disturbance in the air.

[@Fatman305]

Most likely scenario is that the vehicle-to-vehicle protocol, and/or vehicle-to-cloud protocol will replace as needed and add as needed. The environment will look 99.9% identical from 10 cars. If there's interference (even a tall truck in front of you) then obviously you'd want supplemental data from unobstructed vehicles in your vicinity...

[@ristopaasivirta9770]

@@Fatman305 Very true. Vehicles that are able to communicate with each other can greatly increase the accuracy of navigation as well as coordinate with each other to avoid collisions or to dismantle jams in an efficient manner. One should also always try to cover their bases against malicious attacks or injected data. With the possibility of multiple cars sharing information there is a greater possibility for bad data (either on purpose or by accident) to get into the mix. With LIDARS there has been studies on how to perform spoofing attacks against them, but they have not been successful. Against a lot of cameras it is possible to do either "infra-red flashbang" attack if the camera picks up infrared wavelength or targeted laser attack. These can be devastating to the camera that can even result in permanent damage to the device. These attacks won't be noticeable by humans. So this is also a thing that car manufacturers need to take into account in the future. If one would be going around throwing flashbangs then everyone would see, but a pinpoint laser that autotargets vehicle cameras or a infrared burst of light to blind them can be really hard to tell is going on.

[@Fatman305]

@@ristopaasivirta9770 That's another reason not to skimp on sensors (looking at you, Tesla) and some redundancy/physical resilience, as well as "virtual" redundancy, i.e. cm-level live (constantly updated) map. I think it's more likely Tesla will (re-)add sensors, than Google removing sensors.

[@Santor-]

Then they will mutate into a self igniting atomic bomb.

[@dorkarrijohnson4738]

This is so interesting. I wonder if I could actually make something impossible. Or maybe impossible a Model train that has AI.

[@isaacogunmuko3947]

Then whats the point of earning a learners permit or drivers licenses.

[@stephenlester7286]

I am apparently autistic and my brain appears to act like a computer. I was involved in a collision where the court was told the police response was to ask the driver who pulled out at a junction on me whether they wanted to prosecute me. I wished she had said yes!!! So be ready to get arrested in an autonomous car.

[@apurvmj]

How much power does these autonomous driving system consume?

[@m_sedziwoj]

Depending on solution, cameras are few W, for Tesla computer in car is ~75W, for lidars is more and if you use Nvidia it go up to 500W if you use one of they top systems. But only Tesla is thinking about power, because others still working as research and don't care about energy usage, so hard to say what it would like if they start to optimize it.

[@Santor-]

Not much. A triple AAA battery should last 8-14 months.

[@jameshoffman552]

Tesla's FSD doesn't need to identify and label objects in the environment that may pose a collision risk. With its new Occupancy Network, it knows where objects are and can avoid them even if it can't identify them.

[@bobmorr2892]

It's a stupid idea, we don't need self-driving cars, check back with me in a few decades maybe. No computer or machine is ever going to be able to anticipate all the different possible scenarios that a human being can in this situation. Some assistance in certain circumstances from a computer may be helpful but that's about it. Just like a few years ago a self-driving car couldn't tell the difference between the side of a semi truck and the glare of the Sun and somebody got killed because of it. Also some jaywalker in Arizona got ran over and killed. I went around the corner on the back street one day and there was a garbage truck on the wrong side of the road, I'm sure glad I was driving and not a robot. Decades from now if they're like tracks on the freeway and all the cars driving on the freeway have that capability we might be able to kick back on a long drive. Elon Musk and people get this Pie in the Sky BS ideas and people get all excited about it. So this ain't happening in three or four years maybe 30 or 40 or 50 years, get real people. About the only thing this story and others like it are good for is getting views on KZbin.

[@user-it5po2dq9w]

No. Immigration is to be killed so they should invest more into these research so low life losers from third world countries are not to be imported

[@kevinderr440]

I guess I am confused and that is admittedly an ever increasingly easy thing, but regardless of which sensing system is used wouldn’t a simple, “don’t hit shit command,” suffice? So long as an object can be identified that is not open road then the vehicle should slow and prepare to stop until a safe path is determined. I see the problem as trying to design autonomous vehicles to be human and instead we need them to be merely safe.

[@BlakeClass]

Yes, This is exactly what waymo is doing. It could be simplified to “make a detailed map as a baseline of where the immobile shit is, use high powered sensors to map where the new mobile shit is, use high powered computers to predict what the shit is going to do, adjust our behavior based on their predicted behavior, when we aren’t sure what to do, come to a stop until it’s safe to proceed.”

[@PDVism]

Care to explain to us how visual and rader will detect an emergency vehicle approaching and getting the eff out of it's path? In Europe you can be heavily fined and even lose your license if you don't make way for ambulances and firetrucks. Now look at the times that Waymp has been a nuisance to emergency vehicles and that was just in one city.

[@kevinderr440]

@@PDVism I think vehicular interconnectivity is a solution. In the maritime world we already use relatively inexpensive equipment to locate our position and broadcast it via VHF radio. That information is easily recognizable and processable. It may actually improve emergency vehicle access by “forcing” non-emergency vehicles to stop safe and pull to the side.

[@PDVism]

@@kevinderr440 Nice idea, but you actually show us that robotaxis, robocars and robotrucks are far from feasible because it would be much easier to make robo cargo ships because on the sea or rivers there are no obstacles, rarely a blindspot, open flat surfaces, no hills or valleys, and as far as I know not one SINGLE company is looking at making robo ships without crew.

[@golkralifal]

I think they should have initially used advanced sensors combined with image recognition to improve the object detection algorithms.

[@youtubedeletedmyaccountlma2263]

No, they should put up cameras and have a neural network to guide the "self driving cars"

[@chrisoconnell8432]

The problem with image recognition is what happens when the car comes across something it doesn't recognize? Elon Musk has stated that even if a UFO lands in front of the car, the vehicle should still know something is in front of it. Tesla's vision system now transforms its surroundings into a 3D environment so even if it doesn't know what it is, it knows its something and should avoid it.

[@jacobnunya808]

When it comes to mass adoption, you want to keep the sensors as limited as you can. Few want to pay $100,000 dollars for a car just so it can drive itself. That is the benefit to Tesla's approach. At scale good software and processors are likely cheaper than enough sensors to make your car look like an alien spaceship.

[@jamesboston]

Or... and this might sound crazy... trains. Signalized automated metro systems are a thing that exists now and moves millions of people per day at a fraction of the cost.

[@gelinrefira]

Isn't ti crazy that we put so much effort into creating AI for self driving when we can move thousands of people far more efficiently using public transport. Cars have their place but the problem is it has become the primary mode of transportation for many cities.

[@Fatman305]

There's no difference between trains and very sophisticated AI cars that communicate with each other. An endless row of cars can travel at 200mph with 1 inch gap in between them, allow another car to join in middle by slowing to 199.5 then 200.5 then again at 200mph without a scratch...all of them highly energy efficient (solar). Key is in every single moving element on the road (incl pedestrians) being part of p2p network...

[@ChristopherGoggans]

@@Fatman305........ Annnd then a hacker kid decides to spin up a botnet and create a localized DDOS or make a mini portable jammer, and you're right back to square 1 with each car having to be reliant on only itself, it's internal information, and unable to talk to any of the other cars around it. In an ideal world, trains and platoons of 200mph cars all interconnecting everywhere would make many people happy, but there's zero chance of this pipedream happening for the foreseeable future.

[@ChristopherGoggans]

Every comment section I see on self driving videos seems to have at least one person that wants to talk about how trains are the solution to the world's transportation problem! Sadly this isn't the case. It seems that you live either in a city, a nation, or other section of the world, that has the population density, the preexisting railway infrastructure, and the societal impetus and inertia to make trains work for you. If so, great for you and your neighbors, but they are extremely unlikely to achieve the same level of ubiquitous installation and use in the rest of the world. A new network of train tracks being installed for passenger trains that properly covers a city, ala New York City's subway systems is incredibly expensive and has insane capital requirements. No one is realistically considering new subways without some massive and insane changes to government regulations regarding zoning, permitting and huge cash expenditures, with little to no expectations of capital return. Also, it seems to me that most people that are advocating for trains in this fashion, seem to think that there aren't already extensive rail networks all across the continental USofA. There are many many rail lines crisscrossing the country, and there are I think 4 main companies that own and operate the rail lines. Much of the non time sensitive freight is transferred from its origin to a shipping yard near its destination, and a local truck is uesd to pick up the shipping container and move it to its final destination. As with almost any major product or resource, the buyers and sellers will negotiate to find the shipping method that best matches both parties needs. For a lot of the goods in the US, part of that transport is rail.

[@gelinrefira]

@@ChristopherGoggans Everything you said is excuse for not building the correct kind of cities and wanting to keep living a wasteful and wrong kind of lifestyles. Not wanting to change and insist on doing the wrong things just because you want it, is peak America, and is at the core of everything wrong with American culture.

[@balaclavabob001]

Self driving cars are right now are a waste of time. You can't completely check out and fall asleep or read a book or play games on your lap top etc. so you sit there , doing nothing but yet still have to be aware of what's going on. That's like taking a buss ride with a tarantula in a box and the box has a loose lid. That's worse than actually driving yourself .

[@support2587]

How long before these companies beginning lobbying to ban meat driven vehicles?

[@willdehne1]

I agree that the hype of Autopilot and FSD were good to attract Venture Capital and not much else. I have a 2022 Lexus and its driver assistance is very useful for me as long as I remember the word "assistance". I can drive for many hours behind fast moving trucks in any kind of weather. I had a medical Covid experience were I had to rush my wife home, non stop, from Chicago to central Florida. I am 81 years old and can not do that without driver assistance. Let us not forget this benefit for the elderly and keep improving the assistance.

[@sking2173]

If you’re 81, and from the US, VC should mean “Viet Cong” to you …

[@willdehne1]

@@sking2173 No, "VC" is Venture Capital. Did not think of Viet Cong. Were in Germany and wife had had a child as we arrived in 1963.

[@willdehne1]

@@sking2173 "VC" for Venture Capital. Corrected.

[@rajTrondhjem10]

Wow..