Will Tesla be able to deliver FSD with HW3.0 and current Model 3 sensor suite, ever?

[Spooka]

The more I see how slow Tesla is at delivering new FSD features, the more I think it will never happen. Assuming current HW3.0 and sensor suite.
?

No, FSD is already here and running. Elon said so when he PT Barnum’d thousands of dollars from Tesla owners about a year ago who purchased his FSD vaporware. As he grabbed their cash, he promised FSD by the end of ‘19. I believe Elon and every lie he spews so you must be wrong.

 

[NOLADriver]

geofencing is scalable version of your idea. it means that whatever path is within a certain area makes it valid (with probably additional ways to mark something as impossible, but again part of geofencing). this is just better way to do what you are saying. Anything like you suggested (make it path-specific) will not scale at all.

Cool story bro. 95% of the members here can define geofencing. Thanks though.

You are just not understanding. Moving on.

 

[DS369]

Right now the main challenge is DATA. Not the ability to get the 3D representation of the world around the car. Even drones like Skydio 2 can easily see and avoid obstacles while following you. The real problem is this drone doesn't care about what it sees. Car needs not only to be able to know what it sees but also to know how it works so it can analyze and predict the future state. As it appears right now Tesla (HW3) can't even see everything, as in identity all the objects in FOV. Hence the re-write. Example: When a you are on AP and a car is turning left in front of you (coming from opposite direction). Tesla breaks hard as if it's expecting the object to be standing. There is no prediction going on. Normally a human would predict that the car will be out of the way by the time you get there. So we are at that stage, baby is learning the names and meaning of the objects. Also we are not sure if this baby is potentially smart enough to be driving by itself. Meaning does it have the brain capacity? I have big doubts that HW3 will have the capacity to analyze. It might be able to identify X% of objects in FOV with majority of edge cases learnt from the fleet.

Bottom line: Very unlikely that FSD will happen this decade and with this hardware. But it's technically possible when tech and FSD programmers evolve.

 

[NOLADriver]

There are three big hardware hurdles I see Tesla facing: the lack of side visibility at long range, the view distance of the low res cameras, and the current road infrastructure not supporting autonomous driving in any way.

The lack of side visibility really worries me, currently it seems mostly the job of the forward facing wide angle lens, perhaps supported by B pillar cameras that seem really aimed down for road line viewing.

Currently, the camera footage looks like it has the quality of a ten year old webcam. There is no way it can compete in clarity with the vision of someone with 20/20 vision. When I'm driving I'm identifying hazards such as police speed traps, debris on the road, potholes, erratic drivers, etc from a long distance, sometimes up to a kilometer or more. I believe autopilot has a stated range of about a quarter that, 250 meters. At only 120kmph (around 75mph), you cover 33 meters a second. Only about 7.5 seconds to identify and react to a problem (assuming the problem isn't coming head on at the same speed, in which case, halve that). The radar can help in a forwards facing issue, but for monitoring blind spots, it isn't so useful.

Finally, and this is a big one, our roads and driving system as a whole are around the idea of human drivers. Stoplights could have something like infared flashers to send signals to autonomous cars, but they don't and probably won't for decades. Similar tech is used in my city to allow emergency vehicles to communicate with a light ahead to make all 4 directions red and stop traffic. It's not new or revolutionary, but it is expensive to add in and considering no cars use it right now, it would be a waste. Heck, forget that fancy tech, in many parts of Canada they can't even keep lines painted most of the year because of snow.

Another thing to consider, almost everyone in society is okay with 40,000 deaths or so in America because of driving accidents. Now look what happens when autopilot has a single accident with the driver misusing it. International news. This mindset will also take decades to overcome and allow common sense legislation regarding level 5 autonomy.

To sum it up, I don't expect robotaxi's in 2020

What you are talking about with the emergency vehicles communicating with traffic lights is not as high tech as you believe. Its called preemption and it is not a back and forth system. The emergency vehicle has incorporated into its emergency lighting a specific light that is either IR or visible and it flashes at a specific rate and pattern. The receiver on top of the traffic control devices "sees" it and then instructs the timing devices to immediately go to zero time left for all lights to change to red except for the lights controlling the lanes the incoming emergency vehicle is in, which is all green. Its really just a flasher and relay system. One way communication only, and very cheap.

You can buy the same preemption flasher device on eBay. Illegal to use just about everywhere. And the flasher, IR or not, can be seen on the traffic cams. Cool idea if your plate cant be read, and you get caught. Gets you right thru all the red lights.

 

[Sparkeur]

I'm in France and took delivery last week. Very disappointed. Stop signs, roundabouts, turn off at exit on autopilot nav..none working. Also, despite holding the steering wheel as if I'm strangling it, I'm required to move it slightly every 10 seconds. Uncomfortable and annoying.
The rest of the car is fine but I feel I've been blackmailed into enhanced autopilot

 

[erb2000]

The more I see how slow Tesla is at delivering new FSD features, the more I think it will never happen. Assuming current HW3.0 and sensor suite.

I mean - Ok, stop signs. Great. In 5 years development timeline? How long it will be before Tesla releases unprotected left turns on intersections? Another 5 years?

Bottom line - my prediction is Tesla will never be able to release FSD, even in USA, on HW3.0 and current sensor suite.

Now I guess 99% of the forum would disagree with me. I’m curious - is there anyone else who agrees with my prediction?

I agree that FSD will not happen. I've been driving a Model 3 since Thanksgiving and what it does is pretty amazing. I use auto-steer all of them time. The car drives itself 95% I'd guess. But I can't see it ever crossing the double yellow line to pass a bicyclist, or to avoid a parked delivery van. Not to mention the contrived cases that do happen, but are rare (e.g. police directing traffic for whatever reason).

I would be happier if they changed their approach and would have the car learn my daily commute so that it could drive it itself. The car could learn so much from me, but it just doesn't listen.

 

[Svandevo]

The more I see how slow Tesla is at delivering new FSD features, the more I think it will never happen. Assuming current HW3.0 and sensor suite.

I mean - Ok, stop signs. Great. In 5 years development timeline? How long it will be before Tesla releases unprotected left turns on intersections? Another 5 years?

Bottom line - my prediction is Tesla will never be able to release FSD, even in USA, on HW3.0 and current sensor suite.

Now I guess 99% of the forum would disagree with me. I’m curious - is there anyone else who agrees with my prediction?

 

[tbee]

Recently there was a talk about the FSD software, and it turns out that even though object recognition is done using AI, a lot of other things (most notably driving decisions) are done using standard software. They call that 1.0 software and AI is version 2.0. This to me at least partially explains the sometimes erratic, uncertain, and tardy driving style. Rules are by nature very black-and-white, while driving requires more of a soft hand. So once they have the whole system running on AI, I expect better behavior. But no, FSD will not happen in this decade, if ever, it simply is too complex. But highway and bigger country roads should be feasible.

 

[nbergma]

Keep in mind that there is already at least one trucking company in Arizona that has been delivering freight using self-driving trucks along I-10 since May 2019. The company, in partnership with UPS, uses Navistar trucks and the company's own self-driving software. Each truck has nine cameras but also has two LIDAR sensors, which Telsa's do not use. In fact Elon Musk calls LIDAR a fool's errand. Nevertheless, this startup trucking company's end game is to have a vision-based autonomous system which is exactly Tesla's plan.

Now, can Tesla fully achieve this goal with its current hardware, sensors, and software suite? I'm not sure, but news like this makes me hopeful that it can be done.

 

[Svandevo]

I

The more I see how slow Tesla is at delivering new FSD features, the more I think it will never happen. Assuming current HW3.0 and sensor suite.

I mean - Ok, stop signs. Great. In 5 years development timeline? How long it will be before Tesla releases unprotected left turns on intersections? Another 5 years?

Bottom line - my prediction is Tesla will never be able to release FSD, even in USA, on HW3.0 and current sensor suite.

Now I guess 99% of the forum would disagree with me. I’m curious - is there anyone else who agrees with my prediction?

I agree. I’ve been driving nothing but Tesla since 2016. Model S and wife has a 3. Snow builds up on front bumper: AP not available. Heavy rain: AP not available. Condensation in side camera: AP not available. Glare from sun low in sky: AP not available. Unfortunately you also lose all of your safety features like emergency braking etc. I love my car and probably won’t drive anything else. I don’t love the company anymore. Who doesn’t answer the phone and still expects loyal customers?

 

[bobr829]

I believe we are still a ways off. My up to date M3 with 3.0 cannot make a decent 90 degree right hand turn to follow A/P directions on a secondary road. It isn't ready for city FSD driving!

 

[MicDroy]

I agree. I live in Scotland, and our roads are so different from the US. I constantly see situations where a FSD car could never cope. Narrow streets with random obstructions etc. In my opinion, only a human brain is capable of "full" self driving.
I have a 2015 Model S and love autopilot, and am very impressed with the progress Tesla has made.
It's the small number of tricky and unique situations that will be the thorn in Tesla's side.
That said, I also hope I'm wrong!

 

[badsoden]

I concur with the opinion that FSD is not going to happen any time soon, but for a different reason. The sensor suite has dozens or hundreds problems with edge cases, but that's not the real impediment.

1. The car relies on GPS to know where it is, and GPS is only accurate to around 30 feet / 10 meters. That's not accurate enough. The car does not know exactly where it is.

2. The car relies on the map to know where it's supposed to be, and I've personally seen hundreds of map errors where the roads on the map don't match the actual roads. Most of this is in construction areas, but construction areas are a reality -- they exist, and they're not infrequent. The maps don't get updated often enough, in some cases months or years. That's not good enough, you need the map accurately updated in a matter of hours. Therefore, the car doesn't know exactly where it is supposed to be.

If the car doesn't know either where it is or where it is supposed to be, it is impossible for it to do anywhere near full self driving, no matter how good the sensor suite or the algorithms are. You could have the best algorithm and sensor suite on the planet, but if the car thinks it's on the service road instead of the freeway's right-hand lane due to a map or GPS error, you're doomed. (And I've seen both cases, repeatedly).

I see no easy way to solve either issue. There are some hard ways to solve both issues, however, and Elon does like to do things the hard way, so who knows?

I think Tesla has been clear on this. The vehicle control is based on visuals and not on map data or knowledge. Speed signs is a patent issue with MobileEye. Tesla want's to get it done in a way a car is really autonomous and not depending on external information. No 5G, no maps, nothing. It's also clear on Lidar as an too expensive ad-on to normal visual information. The autopilot copies from human driving and you're not driving with a map on your lap. Hence, GPS is required for navigation purpose, but not really for vehicle driving control. This is the real difference from all other efforts for autonomous driving like Waymo, Uber etc.
Nevertheless, sunset blinding is an issue, as well as rain/dirt on cameras. Still wonder how they will solve that. The eye contact with other drivers for communication is an issue but I think it solvable but clearer traffic rules. It's comparable to traffic regulation on bus systems (ethernet e.g.). We must adopt our thinking and traffic society to the new normal. Technically it will be superior and reduce accidents. China will push it and revolutionize logistics with serious costs and time advantages. If we don't follow suit we will have serious economic disadvantages.

 

[matosa]

The more I see how slow Tesla is at delivering new FSD features, the more I think it will never happen. Assuming current HW3.0 and sensor suite.

I mean - Ok, stop signs. Great. In 5 years development timeline? How long it will be before Tesla releases unprotected left turns on intersections? Another 5 years?

Bottom line - my prediction is Tesla will never be able to release FSD, even in USA, on HW3.0 and current sensor suite.

Now I guess 99% of the forum would disagree with me. I’m curious - is there anyone else who agrees with my prediction?

Yeah. It seems to be dragging on but I think the only barrier would be regulatory. Technologically, who else is really anywhere near? Tesla's future business is in Robotaxis, not FSD enthusiasts. FSD now is simply carrots for the masses which helps to fund dev. Any judgement on current speed of progress is purely objective given no one else has ever done it. Refund time for you? I hope not!

Virtue is its own reward. ~ Newman

 

[Jeff Hudson]

This thread is popular and I don't have the time to read every post thus far. I apologize if this has already been brought up.

Over the past few years we all have read a lot about FSD and what people believe. Somewhere else, maybe Electrik, I remember bringing up my belief that Full Self Driving (FSD) [as currently sold] may not be and IMO is not the equivalent to either Level 4 or Level 5 Autonomous Driving. Here are a few of my favorite Autonomous Driving definition links: SAE J3016 automated-driving graphic and Automated Vehicles for Safety I may be wrong but I honestly believe I have never read or heard Tesla claim Tesla FSD would be or claim to be a Level 4 or 5 Autonomous Driving solution. Sure, Tesla's description of the capabilities of FSD appear to be Level 4 or 5 capable and I understand why people might draw this equivalence conclusion because with casual observational thought they seem equivalent. After thinking about this for a long time I think the HW3, programming and peripheral sensor feeds are not capable of Level 4 or 5 Autonomous Driving but the system is capable of FSD with driver involvement and responsibility.

I think where everything went off the tracks is when Tesla brought up the concept of the Tesla RoboTaxi using Model 3. This is a great example of getting the cart before the horse. Of course Tesla would have to produce a FSD system with 99.999% reliability, apply for/receive Level 4/5 Autonomous Driving registration and settle all sorts of legal, regulatory and insurance issues before the first Tesla RoboTaxi general public paying fare. I mentioned RoboTaxi in this comment because I think Tesla may have to address future hardware updates for RoboTaxi to be brought to fruition. And before anyone objects that this is not what they thought they were buying when they checked the FSD option box I invite them to go back and read all of the legal documentation that came with their Tesla Model 3 purchase.

So, what distinction have I been pedantically writing about?

1) Do not confuse FSD with autonomous driving.

2) Tesla will produce a FSD solution with the HW3 computer, programming and present day sensor suite by the end of 2020. Will it be perfect? No. Will Tesla improve upon it with better programming? Absolutely. Is my claim indicative that Tesla will legally be able to use all of the FSD proceeds currently in escrow by the end of 2020? I don't know but I believe this will be the case. Even if it is not the case the realization that Tesla will continue their FSD efforts is comforting and something I look forward to with every software update.

3) One last thing -- read the Owner's Manual -- multiple times!

 

[dramsey]

Well, no. No it won't. And you don't need a highly technical background to come to this conclusion.

All you need to do is look at history.

When I bought my second Model S in December, 2016, Elon was very emphatic that "all Teslas are delivered with all the hardware necessary for full self driving." Not a lot of wiggle room there. He promised a Tesla would drive itself across the country by the end of 2017. They even released a very impressive video showing a Tesla driving itself through the foothills of Palo Alto with no human interaction at all.

We learned that _no_ Tesla delivered then, with what we now call HW 2.0, had "all the hardware" it needed for full self driving. HW 2 cars use an NVIDIA Drive PX single-processor computer. At the time Elon was telling everyone this was all they needed, Nvidia was telling people that a single-CPU Drive PX wasn't capable of this and that you'd need at least the dual-processor version. Elon claimed his amazing software would wring FSD capabilities from a hardware platform the vendor said couldn't do it.

Turns out Nvidia was right. But Tesla had the solution: their own AI driving computer, the "best chip in the world" according to Elon. For those of us who pre-paid for FSD, the computer would be provided as a free upgrade. And at the investor's conference in April of 2019, Elon promised FSD would ABSOLUTELY be complete by the end of the year, and 2020 would see fleets of autonomous Tesla robs-taxis. Buying any car other than a Tesla would be economically foolish, he said.

Well, here we are in May of 2020. What's the situation?

• FSD remains a pipe dream. "Advanced Summon" is at best a party trick, best used in empty parking lots unless you like your car to block traffic when it stops in the middle of a lane, confused by another car, pedestrian, or shopping cart. Apparently the latest FSD software has the fun habit of stopping at green lights. I wouldn't know, because my free FSD upgrade I was promised over a year ago hasn't happened yet.

• In the meantime, Tesla has releases a couple of more fascinating videos showing _just how hard_ it is to really, truly have FSD, and all the super-advanced techniques they're using to conquer the problem.

• Cars with HW 3 have gotten new visualization and functional features that HW 2 cars don't have. That free upgrade we were promised? Not even on the horizon as far as anyone can tell; Tesla certainly isn't talking about it. _IF_ your car has been randomly selected for the opportunity to pay $2,500 for an MCU2 upgrade-- you'll remember Elon poo-poo'd this idea when asked about it-- they'll throw in the HW3 driving computer. You'll lose your radio, of course, because reasons. Not TECHNICAL reasons, mind you. Just...reasons.

So you can talk about fleet learning, neural nets, various AI techniques, sensor suites, hardware, etc. all you want. At the end of the day all we have is many, many years of blown schedules and broken promises. Believing _anything_ Tesla says about FSD these days short of them providing a car for independent testing would be stupid.

What? Oh, I love my 2016 Model S and use Navigate on Autopilot almost every time I drive it. What's your point?

 

[SergeyUser]

Keep in mind that there is already at least one trucking company in Arizona that has been delivering freight using self-driving trucks along I-10 since May 2019. The company, in partnership with UPS, uses Navistar trucks and the company's own self-driving software. Each truck has nine cameras but also has two LIDAR sensors, which Telsa's do not use. In fact Elon Musk calls LIDAR a fool's errand. Nevertheless, this startup trucking company's end game is to have a vision-based autonomous system which is exactly Tesla's plan.

Now, can Tesla fully achieve this goal with its current hardware, sensors, and software suite? I'm not sure, but news like this makes me hopeful that it can be done.

Read this and tell me how it changes your view: The End of Starsky Robotics

 

[badsoden]

I just wonder why trains, especially freight trains, are not yet autonomous? I think that would be the first step to validate FSD.

Trains are running on tracks, they cannot change lane by them self, and use section areas which are the ultimate geo-fence model.

This is the biggest failure of the train systems everywhere in the world. Trains are o dedicated tracks free of any crossing, no steering required and perfect for autonomous traffic. They have no vision to transform this network to a fully automated logistics systems with single trolley for single containers to transport containers fully automated from anywhere to anywhere on the network. The transition from todays train model to such a new model is probably the biggest problem, but the railways are not capable of jumping over the own shadow. Once containers a flowing over streets between (not even into) cities, any cargo per train is dead and past tense. Less traffic means costs per container go up, hence less traffic. Then they will shut down less used lines etc.
They must step away from long trains with one locomotive and exclusive track blocks and move to small, up to one unit self propelling trains/vehicles that are routed through the network automatically and have the intelligence on board to keep distance and and savety standards. It's sad to see this infrastructure being on the wrong track.

 

[AdamMacDon]

What you are talking about with the emergency vehicles communicating with traffic lights is not as high tech as you believe. Its called preemption and it is not a back and forth system. The emergency vehicle has incorporated into its emergency lighting a specific light that is either IR or visible and it flashes at a specific rate and pattern. The receiver on top of the traffic control devices "sees" it and then instructs the timing devices to immediately go to zero time left for all lights to change to red except for the lights controlling the lanes the incoming emergency vehicle is in, which is all green. Its really just a flasher and relay system. One way communication only, and very cheap.

You can buy the same preemption flasher device on eBay. Illegal to use just about everywhere. And the flasher, IR or not, can be seen on the traffic cams. Cool idea if your plate cant be read, and you get caught. Gets you right thru all the red lights.

That's exactly what I'm talking about though, it's NOT a high tech solution, nor does it need to be two way. I'm talking about literally the same idea in reverse (traffic controls flash the car a specific IR pattern for each signal). There is no car to signal comms. Yet such a simple setup could be decades away, nevermind anything more complex. Of course this could make those ebay flashers a lot more dangerous if someone had ill intent!

 

[jaybee60]

I just wonder why trains, especially freight trains, are not yet autonomous? I think that would be the first step to validate FSD.

Trains are running on tracks, they cannot change lane by them self, and use section areas which are the ultimate geo-fence model.

They are but currently only metropolitan services, I believe. The Docklands Light Railway in London has been running largely without drivers for over 30 years now.