Drago Anguelov (Waymo) - MIT Self-Driving Cars

Drago Anguelov is a Principal Scientist at Waymo, developing and applying machine learning methods for autonomous vehicle perception and, more generally, in computer vision and robotics. For more lecture videos on deep learning, reinforcement learning (RL), artificial intelligence (AI & AGI), and podcast conversations, visit our website or follow TensorFlow code tutorials on our GitHub repo.
INFO:
Website: deeplearning.mit.edu
GitHub: github.com/lexfridman/mit-dee...
Playlist: bit.ly/2S1MVdy
OUTLINE:
0:00 - Introduction
0:47 - Background
1:31 - Waymo story (2009 to today)
4:31 - Long tail of events
8:55 - Perception, prediction, and planning
14:54 - Machine learning at scale
26:43 - Addressing the limits of machine learning
29:38 - Large-scale testing
50:51 - Scaling to dozens and hundreds of cities
54:35 - Q&A
CONNECT:
- If you enjoyed this video, please subscribe to this channel.
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridman
- Instagram: / lexfridman

[@lexfridman]

Waymo vehicles have driven over 10 million miles autonomously, which is an incredible accomplishment. I'm excited to see what Drago and the rest of the Waymo team do in 2019.

[@kirkaugustin2232]

Actually, Waymo has not driven any miles autonomously, since all they are doing is following GPS data. They are not capable or reliably detecting curbs, lanes, street name signs, turn signals, brake lites, buses, etc. They can't drive in snow, during solar flares, etc.

[@claus1225]

I want to see them in action in real world traffic.

[@khanhdoquoc7348]

1 year ... and counting.

[@AR-iu7tf]

Thank you very much Alex. This lecture was very informative. Unlike most of the talks on self-driving cars where speaker spends a lot of time on listing challenges and peripherally on how to solve them Drago does the reverse. He shared a lot more with us within the bounds he can divulge which was extremely useful. Thanks again to you and Drago

[@TylerKoz]

Lex, this is a great service that you are providing to the world. Thank you.

[@forgetaboutit123]

This is 10x better than his one on interviews.

[@TylerKoz]

@@forgetaboutit123 They both have their place.

[@ant2011]

Thanks for this. You're really helping people like me get their lives together with these kinds of videos.

[@zakridouh]

These are the type of videos we want, keep it up and thanks :)

[@clauswirnsperger983]

Interesting talk. Shows how complex the problem of autonomous driving is. Thanks Lex for the upload!

[@TomHarrisonJr]

Exceptionally clear, concise and understandable presentation. Thanks!

[@carvalhoribeiro]

Awesome science experiments and ideas. Thank you so much Lex and Drago for lectures like this

[@ritteradam]

The end of the talk about smart agents and learning aggressive / non-agressive behaviour was very interesting, as right now a known flaw with Waymo cars is that they can't take left turns in high traffic. It would have been great if Drago talked more about that specific problem.

[@jialuhan3854]

Highly agree with the Classical vs ML system analogy at 15:51.

[@lennartlut]

Thank you for sharing these ideas, very helpful!

[@KamenRangelov]

Браво Драго ! Ще пусна видеото на сина ми с надеждата да се вдъхнови и да учи повече :-)

[@svetstefanov4904]

Kudos to Drago, WayMo and Haskovo!

[@angelpico3236]

Thanks for the video Lex, would like to see Karpathy doing some similar explanation.

[@PratimaShri]

This is great! Thanks for sharing!

[@kunwang4177]

Wow, learned a lot from this video.

[@cafeliu5401]

Thanks a lot for sharing. I have a practical question: As models grow larger it's harder for learners to follow up with new research. How should we become useful to ai companys without remaking the wheel?

[@guocity]

This is very good,

[@leixun]

*My takeaways:* 1. AutoML for self-driving cars 22:08

[@kargarisaac]

Great talk. Are the slides available?

[@lithostheory]

MIT: 2*2pi = 4pi

[@movax20h]

Talk is nice, but pretty general. The good pieces that are important, is usage of all kind of types of data, automation of training, testing at scale, automated machine learning, and good emulated (but simplified) models of other agents to test edge cases rapidly. Rest is just a lot of hard work, tuning and incremental improvements, that are hard to distill without revealing all kind of secret techniques used. To a big extent the perception modeling is solved, and the rest is just doing more complex modeling and prediction of other agents and special situations involving other agents. Unfortunately a lot of this is kind of circular, so again it is just grown over time to be better and better.

[@ZhengCheng]

nice

[@scenFor109]

Love Anguelov's machine ready definition of reasoning: "Check and enforce consistency of your beliefs. And look for explanations of the world that are consistent." Drago Anguelov Here is one number for The Machine to reason through; #ConcentrationCamps #EndGlobalApartheid Why there is so little data from flying over imaginary lines? And why the military spends so much more maintaining Exclusive Economic Zones than on preventing road accidents.

[@LAURA-if3re]

AWESOME ❤️❤️❤️❤️❤️❤️

[@shivkuma100]

1. How does this simulation + real world data approach generalize to other cities than Phoenix? 2. It appears that Tesla could also do a combination of simulation + real world data... No? What is the incremental value of simulation when you have a billion miles of real-world data vs 10MM miles ? 3. What is the value of the specific package of sensors vs the data set (simulation + real world miles) in terms of effectiveness -- are there metrics of effectiveness -- is it a edge-case driven issue irrespective of metrics?

[@WorldIsWierd]

Tesla already use a combination of simulation and real world data for validation. They call it shadow mode. It checks it's driving while a tesla owner is driving verses their decisions.

[@shivkuma100]

@@WorldIsWierd yes - but that is hypothesis testing. I was referring to creating artificial / synthetic data sets by using simulation scenarios -- how effective is that in creating the right kind of corner cases. Will it actually also have a risk of training the model in the wrong direction if you assume that the simulated data is "real" when in fact it may not be representative of any meaningful corner case ?

[@tomaskadlec756]

@@WorldIsWierd shadow mode is something different, its testing of autopilot, those simulaitons in this video refer to simulate situations that can happen to train autopilot.

[@coronabaytherapy]

Yes. 🤓

[@erikhartley1690]

first comment in a long while, seta heart monitor to the most aware person and when it spikes then set the code to the eye contact of the person and set that to a higher priority and act with s weighted precision

[@MichaelDeeringMHC]

If there are a bunch of waymo cars in the scene with non-waymo cars, will the waymo cars be communicating between themselves and collectively planning actions?

[@coronabaytherapy]

Yes. I would imagine that it will first be a feature and then it will be Mandatory. This network will be absolutely imperative to the future of the autonomous vehicles efficiency. 🤓

[@jacobdavidcunningham1440]

51:53 kind of wondering what's the reason that road was built like that(the squiggle I think in California)

[@qrthrse1]

It's Lombard Street in San Francisco, it's extremely steep. I've been on it, it's touristy. More info on Wikipedia.

[@filmfranz]

So what about Google Recaptcha? Is that data used somehow somewhere?

[@bokoloaranyfa3824]

"Recaptcha" Sure, they made use of it.

[@toonmoene8757]

Pretty interesting as a lifetime (over 55 years of experience) bicyclists trying to anticipate autonomous vehicle behavior ...

[@wongtzerchyuan05]

1:00:12 Didn't know Post Malone goes to MIT.

[@claus1225]

WS value waymo at $200B. what do u think based on these slides?

[@ebzdyberybencjusz]

17:50 so it's basically the same as Tesla, except Tesla has hundrends of thousands of cars collecting data for them. Waymo had to pay for their fleet of vehicles and has to pay their safety drivers, whereas Tesla has customers who do exactly the same not even knowing they train Tesla's NN for free.

[@chris_sndw]

The sample data part is overrated. Tesla aims to make money now with what they have. This project wants to spit out a fully autonomous car when it's ready. That's the difference.

[@bokoloaranyfa3824]

"Tesla has hundrends of thousands of cars collecting data" You have forgotten the small blue icon on the video you have linked, with a man sitting at the computer and labeling the data. Tesla has more data than anyone else, but that is raw data. Most current machine learning needs labelled data, labelled manually by humans. So, it is more important, how many labeller you can pay than how much data you collect. In addition, waymo has lidar, tesla hasn't. Waymo has 3d information ready available for its camera pictures, tesla needs to compute 3d form 2d images, which is self a huge machine learning problem, needing more labeling and more computation. Finally, there is a limit how much data is useful for a given sized neural net. They need much data but there is an amount after which more data will make no difference.

[@user-ni2od5lu6j]

can you share link to original presentation? videos which is inside presentation has low resolution.

[@Pulver]

lol

[@diegodelamota1199]

Go to waymo’s channel and you should be able to find them all

[@lifangu8479]

Some of the critical issues are: whether free will exists, if yes, how it affects human level intuition, and how human level intuition affects car-driving, etc. I design some experiments to verify AlphaGo Zero and other AI technologies do not have human level intuition. The experiments should be arranged very carefully to exclude human intelligence from the AI systems in experiments. So the details of the experiments also should be revealed step by step in a very careful schedule. Based on these issues, it could be proved that SAE level 4 autonomous driving is impossible. It would be a big mistake to ignore these issues and experiments.

[@lifangu8479]

My studies also could explain why the probabilities of some problems would evolve from very low now into very high in future. Of course, there are problems much more severe than free will and human level intuition issues. However, free will and human level intuition issues are enough to prove that SAE level 4 autonomous driving is impossible.

[@lifangu8479]

So Demis Hassabis is wrong, there is no AI technologies which could be a meta-solution to any problem potentially. Intelligence studies also relate to other important judgements. Better logic and experiment criteria are needed. Top research universities better correct those types of mistakes.

[@lifangu8479]

Deepmind scholars summarized their opinions as Reward Is Enough recently, which is wrong actually, because reward is not an effective observable. Short-term rewards and long-term rewards could point to different, or even opposite directions. Things could be more complicated from different situations. I introduce "effective observable" concept in intelligence studies, which could clarify many important issues. Reward Is Enough is what AlphaGo Zero is built on, and relates to self-driving cars. So both AlphaGo Zero and self-driving car technologies have severe problems potentially. Actually the definition of SAE level 4 autonomous driving standard is not good for effective testing, because it does not choose effective observables as the foundation of the definition.

[@codemonkey2k5]

As a Tesla owner, I'm sure you would expect me to bash on Waymo. Don't get me wrong, I love my Tesla and believe that it is currently way ahead of the competition. But that doesn't make what Waymo is doing any less valuable. In fact any time we are working towards a common goal, we are improving. Even if we don't ever reach said goal. That said, I do think that Waymo will get there eventually and each successor their after will have the benefit of it's predecessors lessons learned.

[@paulbyatt3219]

Mr Fridman. I'm 53 and I love driving. I love long drives and driving on long holiday trips, I even like driving down to the shop. Driving is my sole interest. I am a motor enthusiast who enjoys driving my car at any chance I get. I'm a very good driver that concentrates, pays attention and obeys all the road rules. I plan to retire in 7 years and drive around Australia. What will become of my dream? I know there are so many bad and dangerous drivers on the roads, but I believe the vast majority are competent drivers. It's not fair to punish everyone and force them into driverless cars. I see the next generations are going to be very bleak for humans. The way technology is going, there won't be enough jobs for people, and people will not have anything to do for fear of risk. The way I see it, driverless cars is another way governments will control society, we will lose our choices and freedoms. Driving a motor vehicle, to me, is the easiest thing in life to do. We don't need driverless cars.

[@paulbyatt3219]

@infernovoid I understand your point, but we should punish the guilty drivers, not punish everyone. There is an element of risk in everything we do. I see in the future robots will play sport instead of humans, because of risk to injury. Sad future.

[@JK-oc9qc]

i agree with you 90%, we only need driverless car for people who do not have driving capability, or for the last 500 feet of auto parking to save public space and time

[@JK-oc9qc]

as far as jobs, all government needs to do is yo stop stealing from hard working people to give free handout to their voting zombies and stop having more and more regulations. People will figure it out how to make a living themselves, just make yourSelf useful to other people.

[@kumaragurusivasankaran369]

It's interesting that "one model serves all" is followed instead of "one model for one environment". That means there will be single form of AI and less resilient.

[@movax20h]

multiple models each for different environments, are in fact a one model. The fact that there is one model, doesn't mean it doesn't take environment, weather, type of the traffic mix, city, road conditions, into account.The one big model can rapidly adapt, as it changes environment, events happen, etc. It can also determine what kind of drivers are around it. There are two forms of learning at play here. One is the entire model learning, that happens offline, at datacenters, simulations, and replays, etc. That covers a bulk of stuff, including edge cases, and various environments. Then you have a secondary learning, in the vehicle when it drives, and observes environment, these learning is ephemeral and can change from moment to moment, and only covers basic stuff, like other actors characteristics, type of environment, etc. To some extent the second one wouldn't even be considered learning.

[@lotiumdupeuple3661]

Is the blackboard in the back a joke somehow ? :D

[@sheaffer117]

Needs to expand more. Surface area is way ,way too small. Waymo will never see take off if you don't give it a chance too.

[@unRheal]

Even if you get it perfect... Flawless.. There's still that thing about having to pre-program moral decisions when there's a choice about who might die depending on what action is taken.. won't that be tricky?

[@Majenga]

2π x 2 = 4π ... I had no idea :D

[@_PatrickO]

Waymo: it will takes some time to deploy autonomous cars and we cannot predict it Tesla: hold my beer

[@baconwithmusic93]

More like hold my batteries

[@notaturtle561]

Tesla cars are not self driving, far from it. They are highway only autopilot and require constant human supervision.

[@GowthamV07]

@@notaturtle561 Tesla autopilot can do what waymo does easily . Waymo uses walled approach with specific area pre mapped and pre programmed routing , lane changes. Nothing waymo is doing is close to self driving its like programmed driving.

[@thealchemist6355]

@@GowthamV07Whether programmed or dynamic, both approaches are Autonomous as the vehicle drives itself.

[@lifangu8479]

John Krafcik stepped down as the CEO of Waymo. What happened ? Does it work to test driverless cars case by case, without scientific theories as foundation ?

[@lifangu8479]

One type of self-driving has matured for a long time: elevators, which are in well-designed environments. However, car-driving on streets is only regulated, but not in well-designed environments. SAE level 4 definition is not accurate enough for testing. Go gaming is defined in a very small space: 19x19. However, could anyone prove that AlphaGo Zero is an optimal solution for Go gaming ? If not, what are the problems in deep learning and reinforcement learning which make AlphaGo Zero not optimal ? Humans have learning abilities which computers do not have. These problems could cause big troubles to self-driving cars in future, because car-driving is much more complicated than Go gaming. So we need scientific theories to compare the differences in learning-abilities between humans and computers. Without sciences, engineering would not go too far. Without Newton and Cayley's theories, da Vinci was not able to design real airplanes although he was a genius.

[@lifangu8479]

The key is to design experiments with better-controlled conditions. For the well being of students and public, if Alphabet agrees to participate in my experiments to verify whether Deepmind's claim for AlphaGo Zero could be falsified, I could even do them a favor and tell Alphabet one problem of driverless cars. I do not work on driverless cars and have no interests to do business with Alphabet. Having found one problem is good enough to prove my theories. Other people could found much more problems in future if driverless cars are theoretically wrong.

[@frb18neunsechszehn86]

Ich will mein Auto selbst steuern! In der Fliegerei haben die Autopiloten die meisten Erfahrungen. Und die haben es ja leicht: keine Fußgänger, keine Baustellen und Verkehrszeichen. Trotzdem stürzen sie ab. Ja, weil die Mensch-Piloten nicht mehr wissen wie es geht. Die werden nur noch auf die Bedienung der Computer trainiert. (Sully hat das Handwerk noch von der Pieke auf gelernt, nur darum hat er die Notwasserung auf dem Hudson hinbekommen) Das hat schon beim Auto mit der Einführung der Navis begonnen. Die Leute schauten nur aufs Navi, aber dass es zB. statt einer Brücke nur eine Fähre gab, führte sie in den Fluss. Was soll das Vorführen der Versuchsfahrt in Berlin? Nur auf der mittleren Spur fahren, hat der Autopilot Angst mal von der Straße abzukommen? Was hält der Autopilot vom Rechtsfahrgebot? Wenn nichts, dann brauchen wir nur noch einspurige Straßen (wäre wenigstens ein (Park-)Platzgewinn). Und was soll es Ultraschall-, Radar- und Lasersensoren einzusetzen? Wenn dereinst ALLE Autos diese Signale einsetzen, blenden die sich dann gegenseitig, wir wir uns heute schon mit Scheinwerfern? Ein Autofahrer schaut an der Kreuzung einem Anderen ins Auge und schon ist alles klar, der Autopilot such erst mal nach dem richtigen WLAN des Anderen, bis dahin ist der Stau perfekt. Etwa 50 Jahre nach der Einführung von Autos wurde erst der Sicheheitsgurt Pflicht. Was kommt nach 50Jahren Autopilot? Und wieviele sterben vorher? Vom verpassten Fahrspass nicht zu sprechen! ( Ich: 55 Jahre unfallfrei, keine Punkte, und das Auto stand nicht nur in der Garage und noch heute sind mir die meisten zu langsam, außerdem Fluglehrer. )

[@babyrabbit1981]

hi

[@claus1225]

I dont know about you but I dont find this presentation anywhere close to Tesla's presentation.

[@N8N-Soccer]

Roger Lee, good observation. Which do you think was closer to reality?

[@abhaysap]

How self driving car's will run in an country like india where people drive rashly,and there is so much traffic on the road,cars driving neck to neck,and people crossing roads that would be interesting to see.Also when the car is driving on snowy road,or fog how the car is able to see the road,does it treat fog/snow as noise and clears it up and see visibly clear road.

[@codingspace]

Yeah so true whenever he showed a simulation I was imagining how would be the situation in india

[@SahilKhose]

Building an autonomous car in a country like India is a really difficult task. We at Project Manas (a student project in Manipal institute of technology Manipal) aim to solve this exact issue. We are in the last stage of the Mahindra Rise challenge which aims in building the first autonomous car in India. I just got in the team. Hit me up if you are interested in this topic.

[@SahilKhose]

Hey Abhay, if you want to connect find me on any social media platform (IG preferred) by the name Sahil Khose.

[@awocoholic]

they won't

[@kvs1671]

@@SahilKhose any blogs about this project updates?

[@shev1970]

My money is on Tesla. “Literally “ 😉

[@qrthrse1]

I'm convinced that a camera based system will never achieve full autonomy. However, Tesla has cars on the road and will have the best budget Level 2-3 autonomy.

[@DyzioTheDreamer]

A driverless/autonomous vehicle will never ever reliably operate on a public road on which there are ANY human traffic participants. Conversly, driverless cars will become completely reliable once ALL vehicles are driverless. Driverless vehicles will then be able to coordinate their manoeuvres via short distance wireless communication. Very few things can then go wrong. Pedestrians will of course always pose a risk which it will never be possible to eliminate unless their intended moves can somehow be wirelessly automatically coordinated with driverless vehicles in the vicinity.

[@francescodallevedove5018]

Great similarities with Tesla (and probably other companies) work. We'll see who gets there first. For now it seems like Tesla is winning, but who knows. Lidar are used by everyone but Tesla and for know it seems like Elon is right. Anyway amazing work by Waymo, thanks Drago and Lex.

[@unorubbertoe]

Waymo the new dope mule....

[@joecochran5027]

These cars are a, million times better at driving than humans. Unfortunately humans will be sharing the roads with these cars. That will be the biggest obstacle. Interacting with humans successfully. We're stupid.

[@joecochran5027]

@jan simonides don't call me stupid. Explain to me why Im stupid.

[@idumpedinafrenchhorn6811]

@Michael Stubbs AI is estimated to be smarter than humans in 2040, not yet but soon driverless cars will be better and safer than humans. If every car was automated, we wouldn't even need intersections. Also, you dumbass doesn't understand economics. The jobs that are being taken away are being replaced with jobs creating this technology and the cars that use it. So you're minimum wage job as a delivery driver will be replaced by smarter people :)

[@carlknott1081]

Simulation is like practice. Real world driving data is like playing real games. There has never been a championship team or top athletic who got there by only doing a lot of practices. Tesla will win.

[@jerrybender6633]

only 10 mill miles total is slow day for Tesla Waymo then u wanna pay for FSD

[@karlpj1]

Elon said no

[@RobertM-ql4it]

'Waymo | Confidential & Proprietary' - haha....

[@8est8its89]

Having seen this and Elon's autonomous talk... Tesla is at a massive advantage. Their shadow mode alone is enough to whoop ass. Waymo will fail.

[@arielmorandy8189]

29 people gave a negative rating... we should investigate that!

[@sciencetroll6304]

The most difficult part is making sure the human driving the other car always gets blamed when there's a crash. Then the AI doesn't have to do the impossible and read intentions, like we do.

[@alexspera3116]

@Michael Stubbs AI isn't some fantastical concept of sentience as its represented in movies. It's simply a decision tree generated from data and absolutely exists.

[@alexspera3116]

@Michael Stubbs Sentience isn't intelligence, though. Machine learning isn't one of the fastest growing industries because companies want a machine to feel something emotionally. It's a resource to solve problems. Stay away from hollywood for referential sourcing. The Terminator refers to AI in the same way that The Brave Little Toaster refers to a $20 kitchen appliance.

[@PatrickSavalle]

Hahaha, nice. But this guy is clearly still at very beginning of a very, very long journey, no offence. If he thinks the scene at 14 min. is crowded or complex, he should come to Amsterdam, Leiden, Utrecht, Rotterdam etc here in the Netherlands. We won't see any full autonomous cars navigating the Amsterdam inner city in our lifetimes... It's like a 99/1 problem, it's the last 1% that will cost 99% of the effort to get it right.

[@Maggio100]

Lidar could be doomed but can see a white semi in a cross in any light condition.

[@gaelp]

No, radar does

[@darknite84]

I really dont see why Tesla was ranked last in the autonomous ranking. Tesla is having billions of real life miles, and is not constraint to a circuit or a city. Tesla seems to me, to be way ahead of everyone else.

[@francischan57c]

Waymo is useless using lidar. It is not a truely full self driving. It is not a general solution. It cannot drive on roads that is not mapped out. When conditions change, the car breaks down.

[@Seehart]

2:00 I still don't get it. Why make such an ugly car?

[@notaturtle561]

Function focus rather than asthetics.

[@chris_sndw]

It's a study.

[@aus2908]

I wonder if they're not going about it all wrong. Perhaps they should program the car's computer to be able to solve problems on the road the same way they programmed AlphaZero to learn how to place chess without teaching it the rules. There will always be a situation that the car was not programmed to handle.

[@richlinlaw]

i would bet on tesla at least its not boring lolololol

[@oisiaa]

Hmmmm, watching this really makes me doubt Tesla's ability to do FSD. Waymo seems to be WAY far ahead on the simulation front.

[@Martinit0]

That may be, but Tesla has Shadow mode. That is a more natural and scalable way of validating new release candidates. Tesla's pipeline for improving their FSD seems much more scalable and more likely to gather the true crazy corner cases.

[@tbo4338]

Waymo basically said it’s going to take them a long time to be able to drive in all conditions and locations.... Tesla claims their FSD will be ready within a year. Tesla will be first to the finish line, but regulators could probably hold up allowing FSD to the point that Waymo will also be at the finish line by the time it’s permitted everywhere lol

[@toptobottom247]

Dude relax. Ditch the FBI fashion.

[@francischan57c]

Lidar is doomed. Doomed. Waymo please start again or you are doomed.

[@dlwatib]

I found it difficult to understand his accent. I might be able to glean more out of it by viewing it again, but I've already viewed it twice.

[@kirkaugustin2232]

Machine learning requires trial and error, which is totally unacceptable for traffic. Nor can you use things like neural nets in cars because they require large processor arrays in order to work in real time. And the hardware itself is not reliable enough, since there is no manual back up if there is any malfunction.

[@kirkaugustin2232]

Sometimes when there is an emergency vehicle, you can't just stop, but sometimes have to drive out of their way. And other times, the siren is just a civil defense test. There is never going to be an ability to anticipate all the combinations of scenarios. And Waymo has done very poorly. They were kicked out of CA for hundreds of tickets for holding up traffic, going 25 mph in a 50 mph zone, hitting a bus by failing to yield right of way, etc. Clearly this is not a problem that there is a computer solution to. Nor is there any business model, since people spend money on cars because they LIKE to drive.

[@andrewradford3953]

Everything I've read seems to say the opposite. It will just take a little more time for the systems to learn enough to be orders of magnitude better than a human.

[@ZakKohler]

Did your great-grandfather say the same thing about horses? Lol. gosh darn golly these auto mobiles will never catch on everyone loves riding horses...

[@nc3826]

Kirk Augustin Absurd statements such as.... "Nor is there any business model, since people spend money on cars because they LIKE to drive" can't be taken seriously... He needs to go tell Uber and most of the companies on the plant it seems that are developing a mobility service model, that it will never work... It shows he just a troll, mindless repeating false memes....

[@andrewradford3953]

@Michael Stubbs Since cars won't be parked for 90% of the time, there might only be 1/10 of the cars on the road eventually. Also 90% of car parks freed up. Along with real ride sharing. Sad to see the majority of cars currently with only the driver congesting our roads. One day travelling by car might be safer than air travel. Do you demand to control the plane when you fly?

[@nc3826]

@@andrewradford3953 Correction: Since AVs just like "car/ride sharing" are not being parked as much or as long. The cars that are needed to be in service decrease. While the "cars on the road" actually increase. (Just google it and think about whats really happening) Also AVs or any type of caresharing also increase the dead mileage ie the mileage that has be traveled to pick up a customer. So AVs do have negative efficiency issues that have be managed properly to minimize their effects like by using algorithms. And AVs wont initially much of an affect on single use occupancy. Since sharing a vehicle with other passengers is currently rejected by many ppl who will opt due to social preferences.