FSD / AP Rewrite - turning the corner?

[gangzoom]

Yes, it's a step forward, but arguably Waymo are much further along that path. LIDAR isn't that expensive now, either.

Do you think the LIDAR gear on the Waymo Robotaxi cost £100?? I suspect the hardware sensors on that thing is worth more than the yearly earnings of any Uber driver many folds.

It also comes down to the fact no humans need LIDAR to drive (or a HD map of the area), which has always been Elons view, and he is delivering on that.

 

[Glan gluaisne]

Do you think the LIDAR gear on the Waymo Robotaxi cost £100?? I suspect the hardware sensors on that thing is worth more than the yearly earnings of any Uber driver many folds.

It also comes down to the fact no humans need LIDAR to drive (or a HD map of the area), which has always been Elons view, and he is delivering on that.

No, but Waymo are using technology that's several years old, now, as their predecessors had self-driving cars on the streets in 2008, and Waymo were operating self-driving cars on public roads, without safety drivers, in 2017. Since then the price of LIDAR units have plummeted, not because of self-driving cars, but because LIDAR is used in a wide range of commercial devices now, from farm and warehouse equipment to UAVs. Mass production has pushed up the capability of LIDAR and reduced the price at the same time, to the point where I could buy a very capable scanning unit for under £90 last year, and that's now been superseded by a newer model that's cheaper, has better resolution and a longer range. Kids are even using cheap LIDAR modules in robots used in STEM projects.

 

[Durzel]

Let's not delude ourselves that Tesla is infallible.

Tesla resolutely refuses to use a simple rain sensor to control auto wipers, relying on "AI" instead. I don't know how many man hours have gone into approximating what a simple sensor does easily, but you don't have to look far to find people complaining about it. Simple rain sensors are proven tech, work in day or night, aren't affected by occlusions or whatever. Why does Tesla do this? They think they know better.

Whether or not LIDAR is "required" for effective full self driving remains to be seen. I find it hard to believe that having it there, even as an additional sensor, is a "bad" thing if it can be modelled into the cost. From what I know of it the technology is complimentary to cameras, rather than trying to replace them.

 

[LongRanger]

The biggest issue for me on this is the cameras lose visibility at the slightest hint of UK rural or classic UK weather conditions.

Best software in the world dependent on cameras + lack of working cameras = feature incomplete.

All the best hardware in the world has redundant monitoring or backup systems. More cameras doesn’t fix this, better hardware doesn’t fix this. So what is the solution ?

 

[Glan gluaisne]

The external environmental influences on camera capability have to be the weak spot, and the major difference between cameras and the driver's vision. Our eyes are quite remarkable, in that they don't actually have fantastic resolution, yet they are capable of operating in conditions that cameras find too challenging. We also have what amounts to LIDAR built in, as we range find using stereoscopic vision (at least, those of us with two working eyes do). Our eyes are also self-cleaning, as well as being capable of physical movement to reduce the impact of things like glare.

Maybe camera technology can do all that's needed, if it can, then that will be a significant step forward. However, I'm not yet convinced that deliberately choosing a very difficult to implement solution to the FSD problem is necessarily the best way forward. Binning the very cheap and reliable IR rain sensor (costs less than a pound) and choosing to use computationally expensive in conjunction with a camera is a good example of doing something the hard way, just for the sake of it. I know, beyond any doubt, that adding a very cheap IR rain sensor would significantly improve the usability of the wipers, and in turn make me less frustrated when the AI rain sensing, yet again, gets it badly wrong.

 

[Yev000]

So what is the solution ?

Little eyelids on each camera? /s

I think most FSD use cases will involve cleaning/de-fogging the car before use. I never experience blinding once the car is clean and in motion anymore. But sure, getting it to pick you up at heathrow while its all fogged up on your drive during winter will not be possible unless you get your neighbour to go over it with a cloth. By then you should be able to get a robotaxi for a 10th of the fare cost though, so maybe you wouldn't need a car in the first place?

Let's not delude ourselves that Tesla is infallible.

Yeah I'm still frustraded "next turn" info for navigation is not availble to you if someone is changing music at full screen mode. Or not being able to "answer the phone" without touching the screen. I feel these are all transient issues when you are deloping something that drives itself. Pretty sure FSD wont need wipers for the front camera as I never use them on the motorway and it does just fine today.

FYI: when Elon was asked if he would use LIDAR if it was free (yesturday on the Q3 call) he said no, not even if it's free.

 

[Glan gluaisne]

I wonder how much of Elon's conviction on this is because he's forced himself into a corner, years ago, by stating that Tesla would make FSD work just using cameras, plus the forward looking radar and close range ultrasonic sensors? Back when Tesla first announced the self-driving goal, LIDAR was mega expensive, around £100k for a reasonably good 360° unit. Prices have really tumbled, though. Google have kept their sensor development in house, so we don't know how much the Waymo sensor suite costs, but it's estimated to be under £5k. At that price it's not a lot more than the multiple cameras and radar fitted to a Tesla, I suspect, and there's almost certainly room for that price to fall further.

Making LIDAR work over a wide range of conditions is pretty easy, too. Around 35 years ago I made a homebrew directional LIDAR ranging unit for cave surveying, using a salvaged IR fibre optic drive LED and some lenses salvaged from old cameras. That worked reliably out to about 50m, using a retroreflector, even when the lenses were a bit muddy, and inevitably wet (nothing stays dry for long down a cave). That homebrew unit was mainly analogue, in that the ranging system used sine wave modulation of the laser diode, together with a phase measurement system that looked very like two radio IF strips, feeding a comparator. The digital bit was one of the early CMOS versions of the 6800, that measured the pulse width from the comparator, did some simple arithmetic and displayed the distance on a display, with an accuracy of around +/-20mm and a resolution of 100mm (good enough for cave surveying).

The really big advantages of shifting more of the heavy lifting into more capable sensors has to be the reduction in processing that gives, together with the likely reduction in erroneous target ID. The output from a LIDAR sensor, for example, is already pre-processed, in that it gives the dimensions and position in 3D space of everything it sees, with no need to do this using software. The software can then spend more time doing other stuff. This approach probably reduces the power demand, too.

The problem of keeping a LIDAR sensor clean has already been solved for units being used in farm and industrial machines, and farming has to be a more challenging environment for sensors overall. If I had to pick one industry that will see the most rapid growth in autonomous vehicle use then I think it will be farming. There are so many farming activities that are well-suited to autonomous vehicles, and where both costs could be driven down and productivity improved, just by removing the limitations that human operators have, that it's a no-brainer to use autonomous machines for ploughing, soil conditioning, sowing seeds, weed control, fertiliser application, harvesting etc.

 

[MrBadger]

How many lidar equipped cars can you have in close proximity without them picking up spurious signals?

topic drift - there is a time and a place for camera vs lidar, tesla vs waymo and this isn't it imho. This is old tesla vs rewrite tesla.

 

[GeorgeSymonds]

Does anyone remember the rhubarb and custard cartoon character - the videos take me back to my childhood with all the bouncing around

The only thing I've seen that surprised me was the left turn across traffic.and that did seem to be a step forward.

As for LIDAR, its is cheap as chips now - they're sticking sensors in the iphone now. And Musk saying no if it was free - what do you think he'd say with a million cars out there with "All the hardware you need for FSD" and none of them have LIDAR? It would be admitting defeat if he ever said yes.

I agree the price has all this already baked in and the risk is failure to deliver lets just refect on how big that valuation is.

VW is 71B, BMW 41B, Merc 51B, Intel (inc Mobileye) 227B, EDF 31B those 5 companies covering automotive, tech inc self driving and energy add up to about Teslas at 393B

Swap Intel for Waymo at 20B..and Tesla is worth double the value of ALL those companies added together

 

[Glan gluaisne]

How many lidar equipped cars can you have in close proximity without them picking up spurious signals?

No different to the radar units that have been fitted to cars for years, mainly just to feed data into TACC systems. Easy enough to declutter the data, so that only valid returns from the car's own sensor(s) are processed.

 

[Yev000]

I agree the price has all this already baked in

The share price does not have autonomy baked in. They give it the "tech" multiplier and maybe consider a bit of future megapack sales.

Share price, with autonomy worldwide in ~10-15 years would be somewhere aound $2T not $400B. Look at Apple.

But crazy Wallstreet thinking aside, marketcap is almost meaningless these days. Things are only worth how much someone is willing to pay for them. That applies to companies too. Logic has no place here.

You can't use X+Y company is worth less than Z company as a valid argument.

Yeah Ikea bag + Kelvin Clien bag is worth less combined than a Prada bag.... They all hold stuff....

 

[GeorgeSymonds]

The share price does not have autonomy baked in. They give it the "tech" multiplier and maybe consider a bit of future megapack sales.

Share price, with autonomy worldwide in ~10-15 years would be somewhere aound $2T not $400B. Look at Apple.

But crazy Wallstreet thinking aside, marketcap is almost meaningless these days. Things are only worth how much someone is willing to pay for them. That applies to companies too. Logic has no place here.

You can't use X+Y company is worth less than Z company as a valid argument.

Yeah Ikea bag + Kelvin Clien bag is worth less combined than a Prada bag.... They all hold stuff....

You miss the point - The german car companies + Waymo + EDF just dispels the "its an energy company" or "its an autonomy company" or "its just an automotive company" - and you can't just say 1 company doing all those things make it bigger than the sum of the parts. Industry for many years has shown trying to do too many things across a spectrum is rarely a good thing.

Autonomy will be ubiquitous in probably 20 years time, everyone will have it and there will be more than 1 system in operation. At that point nobody will care who's it. The differentiator will not be the self driving part, it will be the comfort and customer service. Its why some people use a bus, some taxi and some use executive transport. Its a means of beng taken from one place to another. And once its ubiquitous it becomes a cost sensitive purchase. The comparison to Apple is wrong for this very reason. People rarely care if its an AMD or Intel processor in their PC, they keep fresh because the produicts keep getting faster and faster and the demands on them keep increasing. Once you can drive sufficiently to be allowed to self drive you're already at the 99.999% reliable place, so where do you go? Its more analogous to sat nav where some loved tomtom, some garmin, now who really cares? 15 years ago you paid 2 grand to have sat nav in your car, its now hard to buy one without it bundled for free.

Thats NOT saying self driving is worthles - it clearly has potential and value, but evenb GCSE economics tells you your car isn't going to go out and earn you $100k a year like Musk has suggested as a robotaxi.

 

[phil4]

so where do you go? Its more analogous to sat nav where some loved tomtom, some garmin, now who really cares? 15 years ago you paid 2 grand to have sat nav in your car, its now hard to buy one without it bundled for free.

You're totally right, but I fear it's even worse than that in the medium to long term. If we forget FSD for a minute, and think more about AP. Tesla aren't the only gig in town, you can now buy other cars that can do just that - ie. lane assist with traffic jam assist and more. It's become so common that EuroNCAP have now started rating on how good it is, and the government is going to do consultation on Auto Lane Keep tech.

Why is that important? Because if those other manufacturers can make something works on the UK motorways and DC's, 99.9999% of the time, doesn't phantom break, get misted cameras, blinded by the sun etc.... people are going to start choosing it, because it'll already have become a generic tech, like the examples.

When making a choice, do you choose a tesla where the AP/FSD works on any road, but not actually very well in all circumstances, or a VW/BMW/Merc/Audi/Jaguar/AN Other that only works on DC's and Motorways but does it nigh on flawlessly?

 

[GeorgeSymonds]

You're totally right, but I fear it's even worse than that in the medium to long term. If we forget FSD for a minute, and think more about AP. Tesla aren't the only gig in town, you can now buy other cars that can do just that - ie. lane assist with traffic jam assist and more. It's become so common that EuroNCAP have now started rating on how good it is, and the government is going to do consultation on Auto Lane Keep tech.

Why is that important? Because if those other manufacturers can make something works on the UK motorways and DC's, 99.9999% of the time, doesn't phantom break, get misted cameras, blinded by the sun etc.... people are going to start choosing it, because it'll already have become a generic tech, like the examples.

When making a choice, do you choose a tesla where the AP/FSD works on any road, but not actually very well in all circumstances, or a VW/BMW/Merc/Audi/Jaguar/AN Other that only works on DC's and Motorways but does it nigh on flawlessly?

Absolutely - my wifes car has BMWs drive assist professional - I think it was a £2.5k option and it is extremely good based on Mobileye eyeQ4 - so the version after the version Tesla used for AP1. They're now starting software roll outs and November it should be getting Traffic light detection (as well as Android Auto - a feature she didn't buy, wan't there when she bought the car but is being added for free), it already does everything EAP did and arguably better and then some, thats the system NCAP reviewed only its better now than the system they reviewed. I think for drive assist the amount I use either the Tesla system in my Model S or her BMW. it does, as you suggest 99.9% of the time, what I want. Above that it would need to be Full hands free, sit in the back and watch a film capability and thats still some way off.

I'll certainly hand it to Tesla for waking up the industry, and Musk has stated thats all he really wanted to do, and they have established themselves as a well known brand now, and who knows we will probably call rapid charging "supercharging" for ever. But Hover will tell you thats not necessarily a guarantee of immortality.

 

[Glan gluaisne]

I can't help but think there is a bit of a conflict between the goal of producing a fully autonomous self-driving car, and the fact that Tesla make some of the fastest production cars on the planet. Those that buy a car based largely on the performance may well not be the same group of people who aren't interested in actually using the performance, those that would rather just sit back and let the car do the driving for them.

I didn't buy the car because of FSD, or autopilot, what ticked the most boxes for me was the range, driver and passenger comfort, that the Model 3 was just about small enough to fit down all the narrow single track lanes around here (we live on one) and as a bonus the performance. I've never used much of the stuff accessible through the screen, other than a play around with some of the toys when I first got the car.

If the car didn't have autopilot it wouldn't bother me, in fact I'd far rather it just had auto windscreen wipers that work properly, as they annoy the hell out of me every time it rains (and sometimes when it doesn't). Where I hope that all the FSD development work does give a potentially very useful spin-off benefit for normal driving is in the situational awareness improvements it might offer. If the sensors and processing become reliable enough to enhance normal driving, by providing easy to interpret information about possible hazards, especially when getting a bit of a shift on, then that would, for me, be something well worth paying extra for.

 

[Dilly]

As someone who recently took the decision to buy FSD, I did so because I like the concept and wanted to be part of the evolution.
If the ability arises, I doubt I will feel the desire to let go of the steering wheel for long. 50 years of steering is hard to let go of!
I will always have an eye in the rear view mirror since by doing so, I was able to get out of the way of being rear ended whilst stationary.
That said, I’m looking forward to it.

 

[Mark-R]

As someone who recently took the decision to buy FSD, I did so because I like the concept and wanted to be part of the evolution.
If the ability arises, I doubt I will feel the desire to let go of the steering wheel for long. 50 years of steering is hard to let go of!
I will always have an eye in the rear view mirror since by doing so, I was able to get out of the way of being rear ended whilst stationary.
That said, I’m looking forward to it.

I'm coming around to the same opinion. The FSD re-write looks to have made a big increase in functionality, and I also want to experience the revolution first hand. I think I will be hitting 'upgrade' on Sunday...

 

[Dilly]

I'm coming around to the same opinion. The FSD re-write looks to have made a big increase in functionality, and I also want to experience the revolution first hand. I think I will be hitting 'upgrade' on Sunday...

I made a number of posts making the point that for me is wasn’t money well spent. In the end, I thought, know what, I’ve spent a shed load of money buying a very fun car and I don’t want to miss a single bit of it’s capability.
Can I justify the cost? Nope. Do I need to? Nope. It’s the cost of a expensive family holiday abroad but will last longer. 2020 - no holiday. No brainier.
I wish I’d been in the same frame of mind when it was £5.8K!

plus, even my wife wanted to buy FSD. My hands were tied

 

[firetrax]

I don’t think Tesla have a lead. The latest system by mobileye (the next generation to the one used by Tesla for AP1) and used by BMW etc in their latest cars is very good. NCAP rated them significantly higher as driver assistance systems.

That was just because of the name rather than actual performance which Tesla AP came out on top on.

 

[firetrax]

Absolutely - my wifes car has BMWs drive assist professional - I think it was a £2.5k option and it is extremely good based on Mobileye eyeQ4 - so the version after the version Tesla used for AP1. They're now starting software roll outs and November it should be getting Traffic light detection (as well as Android Auto - a feature she didn't buy, wan't there when she bought the car but is being added for free), it already does everything EAP did and arguably better and then some, thats the system NCAP reviewed only its better now than the system they reviewed. I think for drive assist the amount I use either the Tesla system in my Model S or her BMW. it does, as you suggest 99.9% of the time, what I want. Above that it would need to be Full hands free, sit in the back and watch a film capability and thats still some way off.

I'll certainly hand it to Tesla for waking up the industry, and Musk has stated thats all he really wanted to do, and they have established themselves as a well known brand now, and who knows we will probably call rapid charging "supercharging" for ever. But Hover will tell you thats not necessarily a guarantee of immortality.

The Tesla NCAP rating was brought down just because of the name otherwise performance wise it came out on top.