EAP and FSD discussion as it relates to Model 3

[eisbock]

which isn't even out yet.

That's my point. I understand designing a car for autonomy when the technology is available.

But building an autonomous car before fully developing the tech? And when nobody really has a clear idea of when the tech will be available? The people estimating level 5's timeframe are throwing darts. Elon has said 2 years for years now. Industry estimates are all over the map. Nobody really knows if/when we'll be there, but the car best suited for that tech is available today. The reality is that the autonomous vehicle will be driven manually until then. How is nobody seeing the issue with this?

 

[JeffK]

That's my point. I understand designing a car for autonomy when the technology is available.

But building an autonomous car before fully developing the tech? And when nobody really has a clear idea of when the tech will be available? The people estimating level 5's timeframe are throwing darts. Elon has said 2 years for years now. Industry estimates are all over the map. Nobody really knows if/when we'll be there, but the car best suited for that tech is available today. The reality is that the autonomous vehicle will be driven manually until then. How is nobody seeing the issue with this?

Cameras and radar have been around for over 100 years... I'd say the tech is pretty well developed.
Are you referring to the upgradable computing platform or the OTA upgradable software?

Oh and I forgot we've been working on self driving for more than 60 years. Did you think this was some crack pot idea out of nowhere that's totally unproven?

 

[Fiver]

Oh and I forgot we've been working on self driving for more than 60 years. Did you think this was some crack pot idea out of nowhere that's totally unproven?

Kind of like flying cars and fusion power? Those are right around the corner also.

 

[eisbock]

Cameras and radar have been around for over 100 years... I'd say the tech is pretty well developed.
Are you referring to the upgradable computing platform or the OTA upgradable software?

Oh and I forgot we've been working on self driving for more than 60 years. Did you think this was some crack pot idea out of nowhere that's totally unproven?

It's certainly developed and has improved a great deal recently, but it's still not there yet. Like I said, I'm sure it'll get there, but as of right now, self-driving doesn't exist. It may be right around the corner, or it may take 50 years when the first self-driving school bus to engulf dozens of screaming children in flames sets back regulations decades. Until then, until some unknown time in the future, we'll be stuck driving an autonomous car manually, which will undoubtedly not be the intended driving experience. It just seems too soon, ya know?

 

[JeffK]

Something tells me that LIDAR may be necessary for the most extreme weather conditions.

But weather makes LIDAR blind, as does mud, road salt, etc.

What benefit for the driving task is achieved through LIDAR that cannot be done with cameras/radar combo?
Your eyes can detect potholes, sticks, small children, animals, cars, signs etc...

Using technologies/techniques like monoSLAM you can make quite accurate 3D maps of your environment with even a single camera and there are eight at various angles.

The one benefit I can see to lidar is that you can "see" in total darkness... but luckily humans don't drive in total darkness and cars have headlights.

 

[JeffK]

I'll be honest, I'm no expert on LIDAR, but considering most of the industry involved in autonomous driving is using LIDAR, including the leader (Google), as well as editorials and criticisms in tech blogs that are questioning Tesla's decision to forgo LIDAR, I'm going to assume LIDAR is superior to Tesla's hardware suite.

LiDAR is superior to just a camera system alone, SpaceX uses LiDAR all the time. Elon's main point was that LiDAR was overkill and not necessary.

I don't think you need LIDAR. I think you can do this all with passive optical and then with maybe one forward RADAR . . . I think that completely solves it without the use of LIDAR. I'm not a big fan of LIDAR, I don't think it makes sense in this context.

 

[eisbock]

LiDAR is superior to just a camera system alone, SpaceX uses LiDAR all the time. Elon's main point was that LiDAR was overkill and not necessary.

But what if he's wrong? Isn't it at all concerning that everybody else in the industry doesn't share his sentiment? He's not a miracle worker has been known to stretch ideas and concepts beyond realistic or practical limits. Hell, he just recently admitted that Model 3's production line will be about the same as current production lines. None of this "alien dreadnought" stuff, just a regular production line with regular technology making cars at regular speeds compared to competitors.

I can see him backpeddling similarly ten years down the road when he introduces AP3 with LIDAR because AP2 isn't cutting it.

 

[vigge50]

I'll be honest, I'm no expert on LIDAR, but considering most of the industry involved in autonomous driving is using LIDAR, including the leader (Google), as well as editorials and criticisms in tech blogs that are questioning Tesla's decision to forgo LIDAR, I'm going to assume LIDAR is superior to Tesla's hardware suite.

Sure, but with that thought you shouldn't think about electric car because nearly all other manufacturer and car blogs told us that electric car wouldn't work then Tesla started to sell there cars. They have prove many wrong and maybe they will do the same thing about LIDAR. I'm not sure myself but I don't believe that Tesla would skip LIDAR and start selling full self drivning if they wasn't highly confident that they could to it.

 

[AnxietyRanger]

Personally I'm still more concerned about lack of 360 radar in AP2 than about LIDAR. Competition does both 360 radar and LIDAR...

 

[eisbock]

Sure, but with that thought you shouldn't think about electric car because nearly all other manufacturer and car blogs told us that electric car wouldn't work then Tesla started to sell there cars. They have prove many wrong and maybe they will do the same thing about LIDAR. I'm not sure myself but I don't believe that Tesla would skip LIDAR and start selling full self drivning if they wasn't highly confident that they could to it.

Cameras/radar is like the Bolt. LIDAR is like Tesla.

Both are electric cars and function practically the same, but one has clear advantages over the other that make it even more practical and useful. You don't need supercharging, but boy does it help. You don't need LIDAR, but your self-driving package won't be as good and it will be harder to develop as quickly as a LIDAR system.

 

[vigge50]

Cameras/radar is like the Bolt. LIDAR is like Tesla.

Both are electric cars and function practically the same, but one has clear advantages over the other that make it even more practical and useful. You don't need supercharging, but boy does it help. You don't need LIDAR, but your self-driving package won't be as good and it will be harder to develop as quickly as a LIDAR system.

I don't agree, if they had LIDAR they would have two option as soon as it started rain heavely or snow, even the system wouldn't work or they would need to develop the system so that it worked without LIDAR. Maybe Teslas system will be harder to code for Tesla but how will that effect me as a user when it's finish(maybe little later)? For the compere with Tesla vs Bolt I don't understand, supercharger makes Tesla more useful in long distance driving with far shorter charging times but if Tesla can manage to do the system without LIDAR I don't see what I as a user would have for advantage of LIDAR.

 

[McHoffa]

If he's wrong he's wrong and he'll admit it, and move on. He just saying that right now he doesn't think it's necessary. They are still testing it and collecting data, there is every chance that they'll analyze the data and determine that they do need something else. That's why they do the testing and don't just ship the cars out and say "ok, we're done".

Except they are selling the fully self driving option now. If it ended up not working with current hardware, they will have to refund all the people that paid for that, or pay to somehow add lidar later to existing cars.

 

[eisbock]

Except they are selling the fully self driving option now. If it ended up not working with current hardware, they will have to refund all the people that paid for that, or pay to somehow add lidar later to existing cars.

There are already lawsuits regarding the capabilities of EAP on AP2. I wonder when the lawsuits will start piling up for FSD. Did Tesla ever promise a timeframe? Has anybody actually bought that feature yet? What if it never materializes over the lifetime of these owners' cars?

 

[eisbock]

I don't agree, if they had LIDAR they would have two option as soon as it started rain heavely or snow, even the system wouldn't work or they would need to develop the system so that it worked without LIDAR. Maybe Teslas system will be harder to code for Tesla but how will that effect me as a user when it's finish(maybe little later)? For the compere with Tesla vs Bolt I don't understand, supercharger makes Tesla more useful in long distance driving with far shorter charging times but if Tesla can manage to do the system without LIDAR I don't see what I as a user would have for advantage of LIDAR.

Yes, my point is that we don't know if Tesla can do it without LIDAR. Nobody does. But the common consensus is that it either can't be done without LIDAR, or it will be very difficult and time-consuming. Neither of those outcomes aligns with Tesla's future goals.

 

[alseTrick]

Personally I'm still more concerned about lack of 360 radar in AP2 than about LIDAR. Competition does both 360 radar and LIDAR...

Definitely.

While I'm not sure Tesla can/will pull off LIDAR-less autonomy, I'm having trouble seeing how they pull off autonomy with only front-facing radar.

As of now, I do not think HW2 is the final version of autonomy hardware from Tesla, and I'm not all that inclined to purchase EAP at $5k, let alone FSD that may never come on these models.

 

[JeffK]

But the common consensus is that it either can't be done without LIDAR, or it will be very difficult and time-consuming.

Keep in mind MobilEye's approach does not use LiDAR either.

 

[AnxietyRanger]

There are already lawsuits regarding the capabilities of EAP on AP2. I wonder when the lawsuits will start piling up for FSD. Did Tesla ever promise a timeframe? Has anybody actually bought that feature yet? What if it never materializes over the lifetime of these owners' cars?

Personally I think Tesla worded the FSD in a more forgiving manner. Not saying there can't be lawsuits, just saying there are more merits to lawsuits regard EAP because on that I could see actually it being argued Tesla misled from the start with talk of AP1 surpassing in December 2016 (Elon), standard safety features "will come" December 2016 (Tesla.com) and a four-camera, auto-lane changing EAP expected to "complete validation" and roll-out in December 2016 (Design Studio). We can now tell those statements ended up being misleading and false, so depending on what Tesla knew or should have known, a case can be seen made.

That said, if FSD in the basic sense that the car can drive itself in most circumstances with a driver present, never materializes, that is of course a different story. But they have left themselves more outs with full autonomy.

Definitely.

While I'm not sure Tesla can/will pull off LIDAR-less autonomy, I'm having trouble seeing how they pull off autonomy with only front-facing radar.

As of now, I do not think HW2 is the final version of autonomy hardware from Tesla, and I'm not all that inclined to purchase EAP at $5k, let alone FSD that may never come on these models.

I guess we know from Elon's famous "12-18 months" tweet that indeed an upgraded version of the hardware suite is coming at some point. Whether it will simply add more processing power or more cameras or more radars or even LIDAR, we do not know. We do know there has been testing Model X and Model S prototypes with what looks like rear corner radars, but that was back in 2015, so in theory it might have been part of the "AP 1.5" testing that was perhaps expected to ship with Model X originally.

I do choose to believe Tesla is prepared to upgrade the current AP2 FSD optioned cars in some manner, if need be. Probably nothing drastic like addings sensors, but at the very least I can see them swapping out the DrivePX2 board for something new (Tesla's in-house silicon?), if it comes to that, and other minor re-call type of repairs if they'd find out the system needs some adjusting down the road.

 

[Reciprocity]

I do choose to believe Tesla is prepared to upgrade the current AP2 FSD optioned cars in some manner, if need be. Probably nothing drastic like addings sensors, but at the very least I can see them swapping out the DrivePX2 board for something new (Tesla's in-house silicon?), if it comes to that, and other minor re-call type of repairs if they'd find out the system needs some adjusting down the road.

They are very likely to at some point build their own chips. Its really the only way to control the cost while getting the aspects that you require. If you take the DrivePX2 for instance, it has processing for Lidar which I believe Tesla will never use.

Again, I have stated this before and will again, more sensors is not always better. The problems come in when your sensors and vision do not agree on what they see. Which one is right? Have you ever heard a sound that you thought was something and it conflicted with what your eyes where seeing? If you design them to always agree then you have no need for the redundant Lidar. Some redundancy is good, but redundant cameras and GPS is all Tesla needs based on their design and Lidar adds nothing. At some point in this process, Tesla validated that they would not need Lidar. My guess is that this happened well before Oct. 16 when they showed that video and had already designed the hardware into the cars. This is not a trivial task and is not something that happened on Sept. 1st and was done by Sept. 30th. They would have had to have been working with Vision+GPS only system for at least a year before that demo, and maybe longer. Why would Tesla put a Lidar in the car? They dont even use the radar for self driving, as its only used for TACC and AEB as well as bouncing under the car directly in front to sense accidents ahead of that car. More is not always better, better is actually better.

 

[AnxietyRanger]

They are very likely to at some point build their own chips. Its really the only way to control the cost while getting the aspects that you require. If you take the DrivePX2 for instance, it has processing for Lidar which I believe Tesla will never use.

Quite plausible IMO.

Again, I have stated this before and will again, more sensors is not always better. The problems come in when your sensors and vision do not agree on what they see. Which one is right? Have you ever heard a sound that you thought was something and it conflicted with what your eyes where seeing?

I know people keep saying this and I keep disagreeing with them. Not necessarily disagreeing with the difficulty of sensor reconciliation, but the overall idea that AP2 would somehow have the optimal amount of sensors.

First of all, I am not talking about LIDAR necessarily. 360 radar, redundant side and back cameras would be additional sensors in my books. Pivoting, self-cleaning sensors would be similar hardware changes as well.

More sensors does not automatically lead to problems, otherwise AP2 would not basically have four redundant sensors at the front (plus ultrasonics): three cameras and a radar. Clearly there is a reason why all those sensors are there.

But going to the sides and the back, there is no redundancy (unless you count ultrasonics which are of limited help plus camera heating). All it takes is one splash of mud to take out the entire autonomy towards the sides and the rear.

IMO for AP2 Tesla chose the absolute minimum, for cost reasons, that they believe they can do some kind of safer-than-humans FSD on. I'm guessing their third attempt at an AP suite will include more hardware for the corner-cases that are left unsolved by AP2.

 

[Reciprocity]

IMO for AP2 Tesla chose the absolute minimum, for cost reasons, that they believe they can do some kind of safer-than-humans FSD on. I'm guessing their third attempt at an AP suite will include more hardware for the corner-cases that are left unsolved by AP2.

I say the opposite, they put to many in the system because they only had one bite at the apple to get it right. Its almost like people, not you, but people in general look at this like it just came into existence as an idea that Elon had on the toilet while pondering the Fermi Paradox while wondering if we are all living in a simulation and Tesla changed their manufacturing process and bought some hardware from the local PC Club and just bolted it all together and called it good. I would guess that they have been working on this since they put the first forward looking cameras in the car. I mean if you look at how quickly AP2 has come into parity with AP1 shows they have a very high level expertise in the area.