Why It's Game Set Match for Tesla Autonomy 

Nice to see a video about Tesla autopilot by someone who actually understands machine learning :) Keep up the good work.

For a video that was published a couple of days prior the big #TeslaAutonomousDay liverstream, you were spot on everything. Well done :)

Tesla: FSD in 2019 Others: If you wave your foot under your trunk, the door opens automatically

Elon stated in a blog a few days back that the reason they do not do active learning is to prevent the neural net from picking up bad habits from bad driving decisions. Instead any time a user intervenes while autopilot is turned on, that info is put in to a queue where it is analyzed to make sure that a change is actually needed. Then the updates are put in to the next software update. So you are teaching your car, but you wont see it until the next update.

Tesla is 5-10 years ahead of everyone else, or at least I hope they are since I paid for FSD. I guess we will find out tomorrow.

This video has answered most of the questions that I had or didn't have. Nicely done. Thanks a lot!

It is super surprising that a Silicon Valley company ranked as one of the three most desired jobs for software engineers in the world might be ahead of companies like Ford and the German companies at a task that is basically entirely a software programming challenge. Oh wait, actually it is incredibly predictable, and yet nearly every "expert" has failed to see the obvious here and kept predicting Tesla would lose this race. Waymo was the only company with a chance of beating Tesla on the tech side, but they have no capacity to manufacture EVs and this race is one that will be won by whoever can mass produce autonomous EVs, not by whoever can achieve autonomous first.

An Outstanding Overview of Tesla's Auto Pilot and its Autonomous Driving Strategy! Thanks for sharing. Looking forward to more great Tesla Content, in the Future.

The key difference between Tesla and the legacy car companies is that it's CEO, Elon Musk, is an engineer. In other words he is interested in engineering and technology, as opposed to law or accounting. His focus is the technical aspects of the things that his companies are building, everything else is secondary. This is the most important trait of his that makes Tesla successful. But other important traits that he has are that he is on a mission to push mankind forward, not just to enrich himself, and he is a brilliant promoter of his companies.

I would rather GEO-FENCE the entire country whichever country that may be. So, pretty much I'm siding with Elon. Not really caring if other companies fail. Sick of their excuses.

Hey FSG! Amazing video, once again! I really enjoy your insight and your analysis! You really deserve ALOT more subs! Do you have a patreon or anything of that nature?

The best explanation of a gan network at 18:00. Thank you for helping me understand it.

Pretty fascinating to see how close your video was to Tesla's pitch yesterday! The line of argumentation was spot on (maybe they even used your video as a blueprint ;) Keep up the good work

17:34 Tencent works on hacking Tesla: Tencent is a major shareholder of Tesla. They work together on streaming,mapping and security. Basically consider their hacking team working from inside Tesla to help Tesla improve the robustness of their product.

Thank you for this fantastic explanation and breakdown of Tesla's full auto pilot!

The major - and most significant - difference between the CEO's of all the others, and the CEO of Tesla, is that all the others barely know how to turn on an electric kettle, let alone how autonomous systems function. Musk understands all of the complex and sophisticated technologies underpinning the systems, and is perhaps the only person on the planet qualified to discuss the merits and demerits of these systems.

Clear well presented information, thank you. I wish I could afford a Tesla

Anyone watching this should look for the 4 hour long conference held today (4/233/2019). They go into great detail and unless they are flat out lying. They are way further ahead than most probably expected. He's calling for Level 5 full autonomy, take the steering wheel out in less than 18 months.

You mentioned GPS is not that useful? It is certainly not instantaneous data like from sensors, but it is essential for context in a way similar to our own human context when driving. I think this knowledge is what makes Tesla's AP far more aware and confident.

Just ordered an M3 with FSD. Struggled about whether to purchase FSD but decided it was something I wanted to experience and contribute to. I think it is awesome but still in its infancy too. I hope the bureaucratic barriers will be overcome and it will be sanctioned for full use. Thank you for the video!

3:18 - Pause for a moment to take in the scale of what is being shown here.

Teslas are also selfselling.

Having worked directly on deep learning for self-driving cars (in particular perception), nothing here stated about Tesla is particularly convincing that their approach is unique, or more likely to lead to a quick solution to this very hard problem. It's nice that you try to get a bit into the weeds with the actual software, but in reality there are many many more moving parts than just slapping a "neural network" onto the problem. Perception, localization, planning, multi-agent navigation, are all separate problems that require individual attention and focus; a full end-to-end model is really not feasible at this point in time, and won't be until we are able to develop more robust models that can handle generalization appropriately. This brings me to my next point; there isn't really a convincing reason to believe that Tesla has some magic novel approach over industry standard. The problems they're facing are the same as everywhere else, and I wouldn't be so quick to just say the rest of this multi-billion dollar industry is just doing "if-else" statements. Many very intelligent people are applying state-of-the-art ML solutions to these problems, and still the challenge is much harder than first imagined. The reason self-driving cars are more difficult than originally thought is because even though driving relies heavily on first-level perception (things like object detection, basic collision avoidance, things ML has been able to do very well at recently), it also has a very non-trivial cognitive component requiring robust world-modeling, something that the entire ML world is currently struggling with. More data is great to help with generalization, to an extent, but the 99% of the unique scenarios that a car will have to deal with. It becomes exponentially harder to deal with edge cases; 10 billion miles full of highway driving will make a car very good at driving on highways. 10 billion miles full of edge cases will NOT mean that the car is prepared to deal with edge cases, because the number of possible edge cases is orders of magnitude more complex and rich than the standard highway driving case. Even if we could collect orders of magnitude more data, the problem then becomes that current models lack the capacity to represent the richness of all the data, and so will fail to learn from it. Full autonomy is certainly further out than first expected, quite possibly by 5-10 years. It will require a complete re-conceptualization of ML, and the development of several breakthroughs in questions that have plagued us since the 60's (inductive learning, symbolic abstraction, world-modeling). Assisted driving will likely come much sooner, but there will always have to be a human to ensure that edge cases are appropriately dealt with.

One...if not the only video I’ve seen that puts all of the technologies within Tesla into a very informative video. I would imagine that cloud storage is also a big part of their IT infrastructure, which captures the the big data, which is then used to make Tesla’s more autonomous and intelligent while in the road, which leads to safety for all. Nice job !

I wonder if level 4 or 5 autonomous cars are able to detect a body of water like a lake or river (or even just a puddle), you know, just in case I decide to park next to one for camping or fishing.

Why not create a set of linked vehicles to maximize space and just step out at your destination oh I guess that would be a train.

Great visual walkthru to understand a very complex subject. Complex because not all sides are being forthright. You help us look overall and see doen the road, putting current data in context rather than picking something and tossing it at us linearly but doing so in a way that forces us to deal with the "issue," knowlng few are well equipped to do so. I feel insulated now from the random fud on this subject. Than k s to you fir sorting all this stuff out.

My concern is not autopilot which, no doubt, is a race that's already over but repairability from other sources other than Tesla. They need better and faster repair service on their vehicles. Waiting weeks for a repair doesn't cut it in my book. I still enjoyed your video quite a bit!

Brilliant video, had to subscribe before the end! So sad that we're so many years from this incredible work here in Africa :(

Thanks,this is a great video and your information is.well updated.Congrats, now I am subscribing to your channel

I'll take the first system that works well in northern Scandinavia Winter roads seriously. Until then I will stay away completely.

I'd buy the full self driving if I could summon the car from the neighbpuring city where my girlfriend works. Then we could get rid of the second car. She could go to work and if I need to get some groceries or use the car for other things she could send the car home, I couöd run my errands before work and the send it back to her. That would be lovely, the car is sitting unused for 22+ hours a day.

I recommend everyone buys a Tesla Model 3 this year, and also invests at least $2000 in Tesla stock

I don't get why anyone would pay more money to not drive their car. Jeeze, take the bus. I DO see a value in MAKING people purchase self-driving if they are convicted of reckless driving or drunk driving. The FSD should be programmed to stay out of the passing lane.

Here is a question about the valuation of Tesla FSD. After that presentation, the stocked moved up 79 cents (yes probably bad Q to be reported weighing it down short term.) However, what if Ford or GM stated that they can "wake up" by end of next year a fleet of 400,000 vehicles currently on the road with a OTA software update and enable their version of FSD for robotaxi with uber/lyft or whoever they contracted with? What wold the stock do? What would their competitors stock do? Something like this? 1) Either there is a conspiracy by ICE Makers to pull Tesla down 2) No one understood what they were presented at the showcase and they will be blindsided at some point 3) No one believes Tesla is capable of doing these things 4) Tesla is really behind and the ICE Makers are farther ahead4 5) Valuation based on numbers given at the presentation: $300K over 10 years * 2,000,000 future vehicles = $600B (upper end)

Worst researched video I've seen in a while. Sorry but when it comes to autonomous driving even Volvo has more advanced systems. Mixing computer vision and control decisions sounds like someone read Tesla's marketing material and made a video from it. Using more data to train your neural net won't give you any guarantees for getting a better net. But then Mr Musk needs to calm his investors after failing to meet his own promises so often that event the slowest of the "old" manufacturers start catching up - and some of them produce and sell more cars in a few months than Tesla has produced ever.

Lidar is just a bunch of dots with very little definition so does not work in a environment made for human sight. A paper bag and a rock is all the same to lidar... Tesla’s system works by recognising everything in a similar way to humans, with the enormous detail of a camera, colour, shade, light texture etc.

Most people think that driving is so incredible. I guess it can be fun sometimes, but a lot of the times it's tiring. Traffic jams are a serious bitch! They make you want to pull every hair out of your scalp sometimes! I hope Tesla pulls this off and the regulators push it through. The world needs this technology! I want to be driven!

Deep learning is all about data. Tesla's have a large number of vehicles running on the road collecting this information including edge conditions as it is happening to the ai system during this use. Imagine a few thousand cars against 100 thousand vehicles that actually is accumulating this data. Tesla has a big data advantage. This why Elon say that the game is already won by them.

Hardware neural net sensor processing. Some impossible things are soon to be possible.

Why do I think Tesla has the upper leg in the EV market? Tesla has advantages over other EV manufacturers because Elon Musk has engineers from the spaceship industry working on technologies that are not accessible to common car makers! Tesla will ALWAYS be ahead of the curve in anything EV-related, hands down! It will take light years for other EV car makers to catch up with Tesla. If I ever buy an EV, it will DEFINITELY be a Tesla, no doubt about it!

Very informative and well presented. Thanks!

There's reason we don't see any of the major car manufacturers using Tesla's technology patents that where open source. Good faith is defined as " you can have access to our patents if we can have access to your". That's not exact wording.

Currently, the neural net is learning from how people drive. Later, people will need their own neural net to learn from computer ( I am referring to beyond driving. Maybe one day we couldl learn to be civil to each other.

What I saw of existing tesla drivers it likes to fail a lot in cases where lines disappear, cars parked on the side of road or there is any temporary divergence and it doesnt understand speed bumps or traffic lights and intersections are problematic. Surrounding car sensisng is also buggy in some cases.

The chip can't be trained! The same way a brick can't trained! Software, the algorithms are trained, or more correctly are developed with the data the cars collect.

I need some Tesla camera for their stock bc it keeps crashing 🤦‍♂️

Great video mate. Really exciting stuff. Subscribed 👍🏻

Basically what you have been saying is that no other car company has taken self-driving seriously.

As an Engineer I've analyzed this space very closely, from a technical level and at each stage over the last 12 years or so (since I first heard that Google had a "secret" driving program) the possible competitors in the space (including Google) have performed one or more major mistakes along the way to autonomy EXCEPT for Tesla. A major mistake is one that has provided a great slow down in advances or has been a knock out blow for the company, such as: 1) Going lidar only early on, big mistake...as Elon correctly states vision is enough and fusion with radar and ultrasonics makes the car vastly super human in sensory ability, zero need for Lidar. 2) Not using multi sensor fusion early on. (Related to above but fundamentally different) 3) Not committing enough resources (money and Engineers) to algorithms and testing early on. 4) *Underestimating* the speed at which the regulatory environment (world wide) would change. (What happens in the US matters less and less for big profits as China rises). 5) Making stupid software and hardware redundancy mistakes (allowing SPF's ..single points of failure in their designs). 6) Relying on one form of simulation (or having no simulation) for fleet learning. 7) Not having a plan to immediately monetize autonomy. 8) Starting too late with their own SDV car programs...this is most of the car companies and Uber and Lyft. :It's been quite amazing watching Tesla make all the right bets and take the right risks to be where they are today...once the robotaxi service goes online they will immediately start earning a hell of a lot of money. In fact , Tesla is poised to have a revenue explosion of the likes that Amazon gained when AWS was released. Autonomy basically takes a car and turns it into a cow not just for the owner but for the company long after the car is sold, they get this without needing to keep ownership of the car which they don not share with the owners in the purchased state and take on a bit more ownership in the leased state but still pay vastly less for the return on investment than any other car owner without a full fleet of autonomous cars they can farm out for the owners or themselves. It's never been done before and it will open a whole new landscape of revenue. I am very LONG Tesla.

Man you were unlucky to release this right before Tesla dropped a whole bunch of new info at their Autonomy Day

Game Set Match Answered at 20:05 you're welcome

FYI about GPS use. You can use GPS for precise positioning (Carrier phase positioning can provide decimeter - cm accuracy); just because Tesla doesn't use it right now, it doesn't mean it cannot be used or isn't being used. Of course there is limitations to using GPS . It won't work as well in urban Canyons and around large tall obstructions. But in most scenarios and coupled with INS it can be great additional sensor for navigation.

Good video. I'm not sure auto makers actually believe Tesla is behind. I doubt they would mention Tesla if they didn't think they were ahead. I do believe old school auto makers want us to believe Tesla is behind. It's kinda been a theme for the last 80+ years with these companies.

Just because Elon says or believes something does not make it true. Sure, the car will drive itself unless the road stripes change in an intersection or there’s snow on the road or there’s a red light or stop sign or any of a million other normal things happen. To deal with all of those things you need machine learning and 3D maps and cameras and radar and lidar. What Tesla currently has is really good driver assist (level 2). Thinking that they can go from that to L3 or 4 just by crunching data is completely delusional.

LIDAR has problems “seeing” in many of the same situations (fog, smoke, heavy rain, dust & blizzards) our eyes have problems with, while RADAR has very few problems in these same conditions. That makes RADAR a much better supplement to visual inputs (eyes, cameras) than LIDAR.

Model 3 radar is easily obscured by ice and snow. While driving in snow this past winter, the radar would go out frequently so autonomy is not possible in inclement weather, moderate rain, snow, ice conditions. During heavy rain this past week, navigate on autopilot would not work during heavy rain. Regular autopilot would work, 99% of the time, but I had to take over control many times due to poor visibility by the cameras. I applied RainX today, hopefully this will fix the issue. I also had to have my autopilot computer replaced a couple of days ago, they gave me the same old Version 3 hardware that came with my car last year, even though new cars come with the version 4 computer, doesn't make sense.

Exceptional video. Thanks!!

They have a little over 1 billion miles not 10 billion. Still an incredible amount compared to all the others.

I think the future will include Lidar. If not leading autonomous driving then it will at least be an assistant. Moreover, Lidar is in use on cars like the Audi S8 to offer active suspension management, which is like Bose's active suspension system, which can go over bumps without upsetting the car.

Driving played out in my 20s. I like being chauffeured now 😎

Self driving car technologies before manuals gearboxes? Bad priorities.

i've been listening to tesla videos on self driving for weeks but i just realised when you said that "Uber doesn't own the cars..." that Tesla's AI is going to get INSANE amount of data once it has a fleet of fully self-driving cars. it's going to be some truly sci-fi level of AI data.

elon musk tells a lot rly obv bs rly often. There is no way radar + cameras will ever be equally good in mapping the environment as lidar is. The other car producers focus on the long term level 4 and 5 selfdriving which can only be achieved through lidar. And i don't think there will ever be a car that will support level 4 or higher selfdriving capabilities with only cameras and radar. So while Tesla knows that it will take up to 5-10 more years at least till we see consumergrade middleclass lidar selfdriving cars they simply benefit from it and fleece all those vegan fitness silicon valley hipsters off their money as long as there is no better alternative. Question is will Tesla miss the future of selfdriving with lidar because they focus too much on the present or will they put enough effort and resources in the background to develop their own effective lidar system.

Good information. However, I still doubt that we will see "good" driving behaviour in the next 3 years. I drive with my model 3 since 1 month and nearly every time I use autopilot, I have to interrupt and end auto mode. Most of the time, because it suddenly decelerates or even does heavy breaking without any reason. A human would not do that. Although I can always handle the situation, I fear that at some time, the car behind me does not. So I will now not use autopilot anymore. Perhaps it would be good to find a signalling worldwide, that shows other people that this is a self driving car, like put constantly on all 4 blinker lights.

I agree with the Ford guy, "Autonomous vehicles will be geo-fenced." We will get there. But, it will be 2-4 decades. The problem with computers is that they do exactly what you tell them to do, just like the MCAS system on Boeing planes. If a little kid runs out behind a car and is killed by an autonomous car, it was not an accident. The car executed a line of code to kill the little kid. The programmer could have avoided the death by having the car travel at 1 mph every time it passes a parked car. However, he chose not to and is guilty of criminal negligence. Likewise, if a car slips off an icy road and kills the rider, the car was, by definition, illegally traveling at an unsafe speed due to the programmer. The programmer is guilty of criminal negligence and manslaughter since he could have programmed the car to go at the legally safe speed of 1 mph. Since it won't be the programmer who pays, but the rich car company that wrote the code, you can bet every collision will be heavily litigated for big, big dollars. Lawyers can't wait for autonomous cars. They will ask for and get the complete code of the car and search through it line by line. In the meantime, "autopilot" will greatly increase safety by assisting the driver. So, the technology is useful and beneficial, and should be explored and expanded. But, until large areas are geo-fenced, drivers will be needed.

If a pelican takes a mullet sized dump on your front cameras, I bet its up to the driver pretty quick!

Ford over-estimated the arrival of autonomous vehicles. I wonder if the SEC will sue Ford. lol

This means Tesla cars will see missiles fired from drones and use evasive maneuvers to save driver. Awesome!!!!

Just being part of the Team ,Wonderful Tesla Team , We shower U all with Blessings , To your Heart & Health B True D.C.H. out

maybe Tesla can use it's neural nets to improve they're crappy customer service? so many videos on RU-vid about it once they have your money they really don't care anymore

Thanks for this video. Some good info about what they are doing. However I think there is a big disclaimer needed in this video which is to say that if Teslas approach is the right approach then Tesla is best positioned to win with this approach. However if there is another approach that doesn't require mass data input and also doesn't require expensive Lidar then that approach could also come out of left field and surprise everyone. Consider that Teslas challenge is to try to leverage their primary asset which is the install base of their cars. So that might be clouding their judgement in what the best approach should be. Waymo has amazing computer scientists and engineers and that asset might also be clouding their judgement. The actual best approach might not have been invented yet and also might not require either of these immense assets.

Thank you for the informative video, great review!

You know frontseatgamer, you've reported well on the sensors suite. Sooner or later, in a relatively short time, Elon will change the sensor package. Today you have the radar and the sonar which support the cameras. This will no longer be necessary in the long run. Tesla will work completely only with the optical (camera) sensors. These are the eyes of almost every other creature on earth, adapted by nature. Lidar is completely absurd. Radar is very bad. Sonics is also bad. We want our autonomous cars to be able to cope in all natural environments. And as we can see, this system of seeing by eyes has won the race in biological evolution. Elon knows about these deep insights. Translated with www.DeepL.com/Translator

That is about as good as it gets. Great work leading up to the next big reveal.

A singular tracked node (STN) w/3D angularity, real-time updated (fluid) via Simulant (1:1 scale, exact real-world 3d) map. Essentially your car plays a video game within the world space as it's own simulation. Other vehicles within your proximity would ping you their STNs. Even better, add real world, fixed nodes for exact calculations.

how come no one is doing mapping for main highways for self driving they could make billions taxiing people just highway exit to highway exit .. maybe in few year?

Geo-fenced autonomy is so easy I bet you could make it in the 1970s. I bet somebody did.

I believe tesla will be geo-fenced but only because of regulators not because of the actual software.

Great video. I hope Tesla will one day detect and decide on what to do when encountering potholes. My city's road is like swiss cheese, so with all the good stuff that Tesla comes equipped with, we need the pothole detection/avoidance here.

"Pretty deep analysis...." Yes, but it would have been so much better if you'd done more to address the question of why organizations like Navigant Research or Consumer Reports for example have Tesla either last or near the last in developing autonomous driving. Why is that?? Seems like an obvious elephant in the room that needs addressing in a video such as this.

It's cool until your car tells you it doesn't feel like driving one day

Quite a lot of slop comes out of Elon's mouth.

To date Tesla has never delivered either on time or on budget ~ so will it be with this one. Further, the vehicle they revealed will not pass the NHTSA (National Highway Traffic Safety Administration). It needs: 1) proper headlights, 2) proper tail lights, 3) rear view side mirrors, 4) windshield wipers, 5) crash tests; which it will not pass as their are no crumple zones with the exoskeleton. That means the cybermobile will KILL anything it hits; cars and pedestrians alike, and it will also KILL anyone inside, as they (not the vehicle) will absorb the enormous impact. Yes it has a cool factor, but at what cost? Mark my words: the base price will increase along with the proposed production timeline ~ and in the end it will not look the same as it does now. And this is IF it is ever produced.

Really no info about how far Tesla is with its autopilot in this presentation. My feeling is that the progress is very small, so they had to push this new hardware to distract investors from missing promised autonomy. Having huge datasets doesn't help with training neural networks, at least because data is not a bottleneck. Models are bottleneck, take state of art detectors like RCNN or YOLO, they still have too many errors to push this into production, pedestrian detection problem is also considered unsolved, if you look through overview papers on scholar.google.com. Lastest reports about disengagement rates show that all autopilots perform 10x time worse then human: www.theverge.com/2019/2/13/18223356/california-dmv-self-driving-car-disengagement-report-2018

It's def learning.....if not Tesla would not be able to understand what its learning ..., but your right it's not adjusting live but waits for Tesla to allow it to adjust to it's learn.

Tip: When you are making these videos, people don't watch them for a few months after you release them, and then never again. They will be watching this video 1 & 5 & 10 years from now. Do you know how ridiculous and antiquated it will sound when you say "now" or "soon" or "currently" "in the next 12 months" even 1 year from now. Date your stuff. Say "As of Spring 2019...." or "in 2020" etc.

WHY AREN'T THEY DESIGNING AN OPEN STANDARD FOR ELECTRONIC VEHICLE INFORMATION EXCHANGE? THEN TESLAS WOULD EXCHANGE INFORMATION WHEN ONE OF THEM WANTS TO DO SOME KIND OF ACTION, IE. OVERTAKE THE OTHER AND COORDINATE THIS ACTION AMONG ALL CARS AROUND (EVEN FROM OTHER BRANDS)

When you consider the expense, how fragile and ungainly the lidar system is, you can see Tesla’s system has gone down the right street. The lidar systems have gone down the wrong streets.

great video, this is exactly what i was looking for! thanks!

lidar is NOT "extremely expensive". Price going down exponentially. I think the last price I saw was just $200. Just a solid state chip now I think. Driving is a visual task, so cameras would seem to be the way to go to me. Musk says that lidar is a "crutch". I tend to agree, but cheap lidar could just be added for almost nothing if it is useful at all or has any application whatever. I don't see a place for it though.

Competitors have to sell their electric cars before catching Tesla. Version 2 chip was in development for 3 years, Version 3 chip already in second year of development. All Tesla cars do not have Version 2 chip. EDIT: Adding V2 chip to all Tesla cars that pay $7k for FSD.

Misses the mark and 80% is of article is speculation. I am rooting for Elon and Tesla and own a model 3. But the self driving features it has rolled out so far have many, many problems. And they are much simpler tasks than full self driving. Tesla and other companies needs a better path to full self driving. How about self driving closed track races? Duh??! It would be wildly popular to watch and safer way of starting to prove it can work. May the best car win. I mean throw in fake pedestrians, dogs, accidents, competitive drivers whatever and see how the competitors deal with it. My tesla utterly fails at lane changes. It can't do them automatically and when I agree to them it is clumsy. It apparently is not doing risk assessments and predictive behavior based on the driver next to me nudging ever so little towards a lane change (but an experienced driver knows from the situation and the slight movements that the SOB next to me is planning to try to get into my lane). Also machine learning has in NO WAY been proven to work effectively. This type of learning in humans is supplemented with other types of learning - instructions, rules, etc. I don't hear that even discussed with machine learning. Please be realistic.

Volvo will now use Lidar at a cost of $1,000 per car, in cooperation with Luminar. So the cost is now insignificant. And no, it won't have all that junk on the roof, but rather it will be seamlessly integrated in the roof. Yeah, things are changing fast. What you "knew" from last year isn't true anymore.

I like electric cars. I hate the thought of robot cars, because it defeats the freedom of a car. We can use options of chauffeured limo or taxi or bus services. This feeling is why I just sold my Tesla stock as well as related ETFs in withdrawing my support in hopes that Tesla's direction and focus will understand that improving quality of life is good. Monopoly is bad. I do hope others will join me in helping to make this clear. Driving is a recreation, a skill and a freedom. To try to make the world safe means stupid people overrun our planet without natural pitfalls. Oh look, its happened already.

I didn't buy the FSD for Australia not even legal right now so saved money for later when I hope all the bugs are fixed and the price drops as it is only a software update. I have used autopilot quite a bit in my model 3 since getting her Oct 1 2019 only done 4k but it has made myself and the wife jump a few times? I'll be going along a clear road 70-100kph and approaching a junction with no other traffic apart from in rear view and the car will lurch quite violently in auto braking as though to avoid an obstacle, I put this down to shadows or line markings or something else its systems are detecting and have to put in acceleration to overcome so the car coming up my rear doesn't hit me since he too can see the road is clear and would not be expecting me to brake so abruptly, I guess this is what the analysis is for so I hope they get a lot better before I pay for FSD, I do love the 3 more than any car I have owned.

OK ... I live in Montana. How is this going to do when confronted by a Deer, Elk or Bison. Will It stop or just give a warning? I just missed a Deer last night by a couple feet. It was 2400hr with a very light mist.

A year or so ago I watched a Model 3 owner repeatedly drive on autopilot his Tesla around/through a improperly banked curve. He did it 3 times. The first and second time autopilot approached the same curve at an excessive speed and he had to take control and slow his car. The third time autopilot slowed to a safe speed on its own. Did the car "learn" independently? Or was the car blessed with a good AT&T cell tower nearby that allowed a quick upload to the Tesla net servers where the "learning" took place and then the new "learning" was downloaded to the car? You tell me. My guess is that 2 GPS locations are noted with a tag for appropriate speed, which is subsequently uploaded to the servers to communicate to any subsequent Tesla car that passes through those same GPS locations. Tesla cars "learn" independently then "teach" the network.

Got my M3 end July. 3rd day drive Vegas to San Fran. mainly FSD and did great. On way down to SoCal my maps started to act up. Maps were literally spinning, and instead of coastal Highway I was driving in Ocean, instead of straight Highway it suddenly placed me in a 30 mile side street ready for left turn and abruptly reduced my speed down. Point is, GPS plays a role, but cameras come first. Just because it thinks I'm on a side street nav me for an upcoming left turn, it did not turn me suddenly into a divider on the highway. It was super annoying, and I badly wanted to use FSD on way home, it reduced speed which was rather dangerous but it did not drive me into a wall just because GPS thought it should. Neighbor asked what if it glitches... so it did and was still safe. Tesla store and mechanics did not know what to do and never seen that issue. Without an update it resolved itself over next few days. Less spinning to 45 degree map turns to back to normal. No clue what caused it. My thought was GPS module was bad, but guess not.

When discussing the flaws of Tesla's AP, why don't you mention its fatal proclivity to pile into stationary objects at high speed? It certainly leaves a torpedo-sized hole in the Game-Set-Match theory.