Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD - the tech vs the reality

pdk42

Active Member
Jul 17, 2019
1,026
1,005
Leamington
This is a really interesting video:


It's from the investor comms event back in April of this year where they went through the FSD tech. Elon brought along the design leads on both the hardware and the software and they both did excellent presentations. I learned a lot from it. Even if you don't want to listen to the tech itself, it's clear from listening to the tech leaders, and Elon himself, that Tesla are hiring some of the finest minds to crack the self-driving problem and investing a ton of money in it.

The fundamental design approach is based on image processing from visual data (just like us humans). Lidar is fitted to the cars, but it's used only as a secondary, confirmatory piece of data to the visual processing. Elon at one point very forcibly pointed out that anyone trying to use lidar as a primary way of self-driving was doomed to failure. I get this since the real world is so complex that lidar alone couldn't hope to deal with it.

As an example of just how well the neural net tech is working, the section of footage around 1:05:16 shows the FSD computer very successfully predicting the bends on a country road. Tesla take vast quantities of real-world video data (i.e. from real Tesla cars, driven by real drivers, in the real world) and annotate it, by hand (with humans!), to train the machine as to what's actually happening. Then, when you and I are driving, the computer is interpreting the video images in real time against this pre-learned data. To do the necessary computations quickly enough, Tesla have designed a dedicated chip focused entirely on the needs of self driving.

All in all, I think the tech is quite amazing really, and it shows just how much Tesla are investing in the field and (probably) how far ahead they are of the competition.

But then I sat back and reflected on my own experiences of AP - and they are not positive. I've only had the car just over a month and yet I've had the whole gamut of issues that I'm sure you're all familiar with - phantom braking, reduced capability due to obscured cameras (by dirt/condensation/water etc), false emergency lane departure events, false alerts when traffic safely crosses our path, and so on and so on...

I get that more data and more learning will make things better, but what if:

- The cameras and hardware aren't good enough or reliable enough (e.g. our eyes have eye lids to clear dirt, we can rotate our eyeballs/head to concentrate on specific things) - none of which the car's cameras can do.

- Elon's fancy chips are still 1000x too puny to do the job well enough, fast enough.

- The whole neural net idea just lacks the necessary "intelligence" to do the job properly.

These possibilities aren't completely left field. Neural nets are fundamentally just pattern matching to previous data - there's no intelligence behind any of the inferences, just a probability fit (and yes, I know that I'm verging on the philosophy of intelligence and whether "AI" really is intelligence at all!).

Anyhow, random thoughts for a Sunday evening!
 
Last edited:

vitesse

Active Member
Apr 2, 2019
1,021
285
Hertfordshire (UK)
I'm not optimistic either. I haven't watched the video yet but to me the fundamental spec. of the Tesla 'sensory' hardware looks too limited. I do understand that Tesla is doing things differently to its competitors, but that doesn't detract from my concerns.
 

pgkevet

Active Member
Jul 1, 2019
1,134
972
mid wales
Also look at : Andrej Karpathy at PyTorch for additional details re AI

My take on the whole thing is they're going to need a heck of a lot more processing power and a lot more time. Some folkbeliev that the AI trainign will be exponential but personally I believe that Tesla will add one nut at a time to crack.. so 'simple' enough things like road signs, traffic lights then on to more complex junctions, roundabouts, looking further ahead, then finally starting to look at the stuff we analyse without thinking - human behavioural clues, the way cars position themselves and their drivers behave which clues us to who is givign way, taking the right of way etc. Yes they are moving ahead and planning more processing power which isi where any exponential growth will come from but in years, not months - then finally they are going to have to duplicate sensors and design cleaning systems.
 
  • Like
Reactions: Bernard_S

gangzoom

Active Member
May 22, 2014
1,186
997
Uk
Neural nets are fundamentally just pattern matching to previous data - there's no intelligence behind any of the inferences, just a probability fit

This I think is a real kicker that might sink Tesla, the worst thing is Elon has bet the entire future of the company on been able to deliver 'Robocars'.

Who knows maybe Tesla can crack it, but it seems unlikely. Still I was and still am happy to have paid £5000 for FSD. Its not every day you get to take part in such cutting tech advances, and the money is/was forgotten very quickly compared to the anticipation of technological advances.
 

EVMeister

Lover of Tesla
Apr 8, 2018
1,585
10,814
England
This I think is a real kicker that might sink Tesla, the worst thing is Elon has bet the entire future of the company on been able to deliver 'Robocars'.

No, he hasn't. If FSD doesn't happen as he envisions and robo taxis don't become a reality, Tesla still remains a cutting edge sustainable energy and transportation company. Sure there will be a lot of "wasted" time and resources, but it's not pivotal for the company's future survival.
 
  • Like
Reactions: Obliter8

pdk42

Active Member
Jul 17, 2019
1,026
1,005
Leamington
This I think is a real kicker that might sink Tesla, the worst thing is Elon has bet the entire future of the company on been able to deliver 'Robocars'.

Who knows maybe Tesla can crack it, but it seems unlikely. Still I was and still am happy to have paid £5000 for FSD. Its not every day you get to take part in such cutting tech advances, and the money is/was forgotten very quickly compared to the anticipation of technological advances.
I can see your point here. And certainly, after watching this video, I'm prepared to cut the AP some slack. I guess I'm just a techie nerd at heart!
 

Chickenlips

Member
Oct 18, 2019
53
27
Kent
I only collected my car last week, so I haven't had long with it and I don't have FSD, but from what I have seen from AP I think we are a loooong way off from FSD. Maybe it's because I'm only on v9? but I really have found it to be pretty bad. It isn't planning far enough ahead and maybe that is due to poor resolution cameras so I think there is going to have to be at least one more hardware update, if not two. Hopefully it is better with v10 though?
 

Tony Hoyle

Member
May 7, 2019
923
557
Stockport, UK
It's basically lane following with some clever bits at the moment. It gets better/smoother with each release, but nowhere near fast enough for the pronouncements from elon to make any sense. FSD is a pipe dream.
 

pgkevet

Active Member
Jul 1, 2019
1,134
972
mid wales
Often there isn't any. In our country roads snow plows routinely hit them, combines hit them, road mowers hit them, etc.

In the UK we have national speed limits: 70,mph on motorways or dual carriage ways, 60 on all other roads except towns where marked otherwise and/or indicated by such things as evenly spaced street lighting (30mph unless indicated otherwse) Where we have speed limit signs they are followed up by repeater signs so it would take multiple incidents to knock them all down.
But my main argument against the progress of Self Drive is not the markings but it's current failure to look far enough and anticipate how tight a turn is or the next bend ahead. Indeed single track country lanes are officially 60mph even when 20mph would be pushing your luck and if i did (foollishly) put on cruise control then car would try to accelerate. Any more than the system allows for weather conditions.
 

marvin28

Member
Oct 22, 2019
36
26
Sheffield
If he so adamant that visual data is the way ahead why, oh why, does he rely on a GPS database for speed limits?
Other manufacturers have been using camera technology to recognise speed limit signage for years.

Unless an order has been made and the road is signed to the contrary, a 30 mph speed limit applies where there are three or more lamps throwing light on the carriageway and placed not more than 183 metres apart. So it would need to be able to recognise street lamps and measure the distance between them.

GPS is also beneficial if a sign is missing or obscured. Imagine careering around a villlage at 60mph because a sign had been knocked and spun out of alignment.
 

elecTED

Member
Jun 21, 2019
324
320
Bristol, UK
Get the basics right first (autopilot/tacc/auto steer) and FSD can follow one day. (Today FSD is still fake/future self driving).

I’ve seen better implementations of basic tacc and auto-steer in 20k cars. I think Tesla is using a sledge hammer to crack a nut type approach here for basic auto pilot - too many fancy radars, cameras, algorithms etc to keep the car just in between two lines and hence all the dangerous phantom breaking, over-zealous lane departure/crash impact warnings.

The whole self driving thing is over-hyped at the moment. Yes cars can drive themselves today with existing technology. When will Joe Bloggs be able to buy a self driving car? One day quite possibly but not any time soon is my guess.
 

pdk42

Active Member
Jul 17, 2019
1,026
1,005
Leamington
Get the basics right first (autopilot/tacc/auto steer) and FSD can follow one day. (Today FSD is still fake/future self driving).

I’ve seen better implementations of basic tacc and auto-steer in 20k cars. I think Tesla is using a sledge hammer to crack a nut type approach here for basic auto pilot - too many fancy radars, cameras, algorithms etc to keep the car just in between two lines and hence all the dangerous phantom breaking, over-zealous lane departure/crash impact warnings.

The whole self driving thing is over-hyped at the moment. Yes cars can drive themselves today with existing technology. When will Joe Bloggs be able to buy a self driving car? One day quite possibly but not any time soon is my guess.

I agree with pretty much all of that, esp the sledge hammer to crack a nut analogy. It's the first car I've had with anything more than basic cruise control, but I can see that TACC would be really useful. Having recently driven a friend's Golf, whose TACC worked perfectly, it's really frustrating that all this neutral net technology can't get the basics right.
 

Artiste

Member
Jun 17, 2019
405
279
Lancashire
If he so adamant that visual data is the way ahead why, oh why, does he rely on a GPS database for speed limits?
Other manufacturers have been using camera technology to recognise speed limit signage for years.

This really infuriates me about the M3. For several years I have enjoyed traffic sign recognition (courtesy of Audi) that works instantly and flawlessly. Not only does it recognise every speed sign, and show a warning on the head up display if I’m going too fast, it’s also invaluable on endless motorway roadworks. If I’m using adaptive cruise control it will automatically slow down when I enter roadworks, maintain the restricted speed limit, then speed up when I exit the roadworks. It also recognises changing speed limits on “smart” motorways. I’m sure it’s saved me many a speeding ticket, especially on really long stretches of road works that use average speed cameras.

Not only does my M3 do none of this, but it’s GPS database of speed limits is seriously flawed. None of the zillions of 20mph roads near me are recognised, and they’ve been in place for years. Several roads I use regularly change from 40-60 and back again many times and the car hasn’t a clue.

Another example of how buying an apparently incredibly technically advanced car has set me back years. Add that to the crap automatic headlights, the crap automatic windscreen wipers and the not fit for purpose TACC. And Elon said FSD would be “feature complete” by the end of the year. Bollocks!
 

vitesse

Active Member
Apr 2, 2019
1,021
285
Hertfordshire (UK)
In the UK we have national speed limits: 70,mph on motorways or dual carriage ways, 60 on all other roads except towns where marked otherwise and/or indicated by such things as evenly spaced street lighting (30mph unless indicated otherwse) Where we have speed limit signs they are followed up by repeater signs so it would take multiple incidents to knock them all down.
But my main argument against the progress of Self Drive is not the markings but it's current failure to look far enough and anticipate how tight a turn is or the next bend ahead. Indeed single track country lanes are officially 60mph even when 20mph would be pushing your luck and if i did (foollishly) put on cruise control then car would try to accelerate. Any more than the system allows for weather conditions.
In theory, Teslas are data-gatherers for Tesla's virtual world in the Cloud and in future the cars will be able to rely on previously gathered speed information at any turn and approach that turn at an appropriate speed.

This aspect of Tesla's model is fine. What I fear is that surprises will be a major problem - like something small or low enough not to be detected by the radar or the cameras in the middle or the lane.

I also don't believe kerbs are reliably detectable - using a simple camera system means you need lines or uniform and distinct edge of the road detail to provide enough data for safe lane-centring. Depth-sensing ToF (Time of Flight) camera technology has arrived but I'm not aware it's used by Tesla yet. I know my car gets nervous and can give up on lane keeping when it rains hard.
 

vitesse

Active Member
Apr 2, 2019
1,021
285
Hertfordshire (UK)
I agree with pretty much all of that, esp the sledge hammer to crack a nut analogy. It's the first car I've had with anything more than basic cruise control, but I can see that TACC would be really useful. Having recently driven a friend's Golf, whose TACC worked perfectly, it's really frustrating that all this neutral net technology can't get the basics right.
indeed, AP1 (MobileEye) TACC is actually very good and much more refined than an AP2 S75 loaner I had for a week a couple of months ago. But owners do report Tesla is gradually improving.
 

vitesse

Active Member
Apr 2, 2019
1,021
285
Hertfordshire (UK)
If he so adamant that visual data is the way ahead why, oh why, does he rely on a GPS database for speed limits?
Other manufacturers have been using camera technology to recognise speed limit signage for years.
AP1 does read road speed signs and rather well it has to be said. But you need a system that has both sign recognition and GPS linked to a cloud database because road signs can be wrong. Only yesterday driving along the A3 into central London a 30MPH sign that should have been facing towards traffic approaching from the left was erroneously turned so the car could see it and it recognised it and recommended an incorrect maximum of 30 when the actual limit was 40 at that point. I've seen bigger discrepancies.
 

About Us

Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.

Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.


SUPPORT TMC
Top