UK - Transitioning to Tesla Vision - removal of Radar

[Strategium]

Moderator comment - This thread merged from various other threads and posts including "'RADAR' = 'Removed And Driving Autonomously Regardless'?", Phantom braking getting bad. (TACC/AP) [Not AEB] and The Truth About FSD?

Please keep this thread UK and Ireland centric

Latest comments from Elon Musk seem to confirm that he is now determined to remove RADAR from the Tesla fleet and solve FSD using vision alone. I fear this means even more delays to when we can expect to receive some form of usable FSD system if the Tesla software team have now been sent off to rework the whole FSD implementation yet again (son of rewrite!?) to now ignore RADAR input. Is this the reason for the lack of any new updates to the FSD beta in the US? Is RADAR the reason why Tesla hasn't been able to cure phantom braking?
Am I the only one wondering why I had to pay Tesla to put a RADAR in my car that apparently they don't need after all and will switch off in future?

 

[NewbieT]

He’s been a vision advocate for a while. I bet radar is the cause of at least some phantom braking but shame they can’t integrate the data better. e.g. Fog= use radar. Camera blinded= check radar.

 

[NewbieT]

Another thought. Humans use different senses for different things. We listen and look for traffic before we cross the road.

If they’re so convinced about binning radar, why do the R&D for their own RADAR (unless it’s that that has convinced them it’s not the way forward?)

https://twitter.com/x/status/1319363510011383808

 

[Mrklaw]

With stereo front cameras they should be able to calculate distance fairly well, but that would have limitations at far distance?

 

[gangzoom]

Another thought. Humans use different senses for different things. We listen and look for traffic before we cross the road.
]

But not radar, and when driving listening doesn't help much, its all vision based.

Extra sensors isn't whats going to enable autonomous driving, but at the same time am not sure just trainings NNs off massive datasets will either.

As far as I know the current NNs don't 'learn' by themselves. All of us know the complex/difficult junctions near us, we known when traffic builds up in certain lanes, etc. Current AI deployment will never do what humans can do with adapting on the go.

My understanding is current NNs are essentially doing complex regression/risk analysis based on past behaviour from the training data rather than creating new ideas on how to solve problems.

I was convinced autonomous driving is coming soon, but working with a few AI 'experts' recently on a different area leaves me far from convinced current AI can deliver. The problem is NNs really aren't 'intelligent' at all, they are infact the opposite, dumb and only good for following protocol.

So the real question I think is do you need 'intelligence' to drive.......if so than the current approach by everyone is doomed. If not, than there is a chance.

 

[NewbieT]

NN = pre programmed algorathim that you can’t audit the code for (eek). They do simulate the brain. Different paths with different weightings. Many stimulus/inputs affecting the outcome/decision. Agree, each car isn’t dynamically learning (way to risky due to unpredictable results). They do send back fringe cases to be added to the training data sets in the dojo which develop later version NNs. Interesting when you change and add to the training dats set, you can actually make the desired result worse. Means choosing trading data carefully and lots of iterations to get it right.

Humans I think use much more memory when driving, the line around the same pot hole every time. Which lane to be in. Tesla seem to want to be purists and drive every road like a Tesla car has never seen it before. Humans drive differently on a road they haven’t see or when they detect a change, or the traffic is bad. AP just seems to have one mode of driving and doesn’t seem to adapt to the situation. I have seen the car leave a bigger gap in the rain though. My point: on AP were all braking or taking over at the same places, Tesla should use theirs flags to see why and change the fleet speed. There is not point speeding head long into a known hotspot.

 

[phil4]

It's an interesting issue. I agree that NN could be boiled down to a set of rules, it's just a better way to write those rules... but they're still just a set of rules.

In that regard, is it really AI... as in intelligence? It will do what the NN says for each and every input scenario... it won't reason, and change that. So is it intelligent.

That being said, that doesn't mean it can't work for the majority of driving. After all for most people most of the time driving is reacting to the inputs we see/hear/feel.

I'd also suggest that the solution that's arrived at doesn't need to be "better" than a human, just safer. It'll also need to be not totally alien to the passenger (so no Ken Block escapades), and not much slower than both expected from the passengers, but also other cars on the road. After all, how do you define better? How do you even see it's covered the brake as it's anticipating a bus around the next corner because it spotted a freshly stubbed cigarette at the bus stop it just passed? You don't, and perhaps it doesn't need it.

 

[Strategium]

Reply to Moderator comment - unique focus below on Elon's recent tweets & comments on Tesla FSD progress for UK & Ireland, featuring forum-specific 'British sense of humour'

Elon Musk, an FSD beta tester and a journalist walk into a bar, sometime in 2025.
Elon Musk says loudly, 'We have a HUGE new upgrade coming to Tesla FSD. It is almost ready - we haven't just ditched LIDAR, RADAR & ultrasonics, we are now completely getting rid of all cameras! All in on 'human vision' which is the best way to achieve autonomous driving because we now realise humans are able to do this driving thing with no external technological props and crutches."
Turning to beta tester, Elon adds - "We just need to implant our neuralink hardware 5.0 chip in each driver's brain so they will control the car directly with their mind."
Journalist says, "Wow, Elon that's amazing, how long will be it be until beta tester here gets to try it?"
Elon replies, "Oh, about 2 weeks"
Journalist says, "Two weeks! Really?"
Elon replies, "Yes, definitely, I have been driving with an implant myself and simply cannot see how this won't be ready for roll out in 2 weeks."
4 Weeks later... "Of course this was only aspirational because it is a very hard problem and in any case won't be available for Tesla FSD drivers in UK"

 

[MrBadger]

Whilst not UK relevant, yet, but a heads up that future new Model 3 and Y for North American market will be delivered without Radar and some current safety features temporarily disabled or limited. Strikes me as the point of no return.

Tesla.com - "Transitioning to Tesla Vision"

https://www.tesla.com/support/transitioning-tesla-vision Transitioning to Tesla Vision We are continuing the transition to Tesla Vision, our camera-based Autopilot system. Beginning with deliveries in May 2021, Model 3 and Model Y vehicles built for the North American market will no longer be...

teslamotorsclub.com

 

[pgkevet]

So no radar, no gear shift, take away half the steering wheel..might as well lose the scrolls and do it all by voice "Brake, BRAKE" ...
Any truth in the rumour of a 3-wheeler, passenger seats optional extras?

 

[NorfolkMustard]

Very odd timing for this. Why not wait for the software to be ready rather than having to apologise for loss of functionality.

 

[Dilly]

So no radar, no gear shift, take away half the steering wheel..might as well lose the scrolls and do it all by voice "Brake, BRAKE" ...
Any truth in the rumour of a 3-wheeler, passenger seats optional extras?

So, basically it will become a four door, five (or seven) seat, home cinema. Premium connectivity permitting…

 

[Dilly]

Very odd timing for this. Why not wait for the software to be ready rather than having to apologise for loss of functionality.

The apology is easier at the outset. No need to apologise again during the coming weeks/months/years*. (*delete as appropriate)

 

[Imrie]

Very odd timing for this. Why not wait for the software to be ready rather than having to apologise for loss of functionality.

possible due to component shortages...accelerating this plan

 

[gangzoom]

Very odd timing for this. Why not wait for the software to be ready rather than having to apologise for loss of functionality.

Its the way Tesla works, for Elon going 'all in' seems to be the only option in everything. Given his track record with Tesla and SpaceX I wouldn't bet against him.

 

[Yev000]

I was convinced autonomous driving is coming soon, but working with a few AI 'experts' recently on a different area leaves me far from convinced current AI can deliver. The problem is NNs really aren't 'intelligent' at all, they are infact the opposite, dumb and only good for following protocol.

Ok I think we need to step back and take a breath. There is no need to assign human intelligence metrics to NNs or any AI for that matter.

By definition it was not possible (given impracticalities) to manually create an NN before 2017.

I am not an 'AI expert' by any stretch of the imagination, but I have have 2 takeaways:

1: It is not possible (within a reasonable time limit) to manually program a car to drive itself without intervention *everywhere*. People have attempted
2: NNs make it possible to shrink the coding period by automating *writing* the code for millions of cases.

We should not call it "dumb" just because its not an GAI capable of philosophy.... My "smart"phone is "dumb" by that definition.


The magic of all this is that no-one need be convinced... One day it will happen and will require zero convincing.

Also, don't buy any product based on future promises, including Teslas...

 

[Rustybkts]

If the radar is not going to be used at all, how will the emergency braking work that is supposed to stop you driving over pedestrians in wintry days when the windscreen clears but the camera's do not as they rely on a poor heater on the windscreen.
Mine used to take forever with the wipers flat out because the AI if it is really there thought it was raining hard.
Will the safety system that reduces the accelerator when facing a wall also not work when the windscreen is frosty?

Surely the radar has some uses?

 

[mpandrew]

I vaguely remember seeing videos of the radar bouncing underneath the car in front and detecting a car stopping in front of that, essentially beating the visual system. Seems odd but this is the the brave new world of being Tesla beta testers!

 

[Medved_77]

Chris aka @DirtyT3sla did an excellent video on the removal of radar, what it means for FSD beta, how it will impact new US deliveries and the features that will be disabled pending a future software update.

 

[tomorrowman]

Moderator comment - various posts moved from The Truth About FSD?

Interesting presentation from head of Tesla AI (Andrej Karpathy) regarding Tesla's move to vision only FSD.

Bridges and phantom braking featured

Worth a watch if you have both the time and patience and I'll let you draw your own conclusions but it left me a little more optimistic

I still struggle to see how it will manage the 4 lanes (with barely visible lane markings) 'roundabout' of confusion known as the Basingstoke Black Dam roundabout - Christ humans can't even navigate it never mind computers.

Can see the AI is going to have some serious learning to do

No timescales mentioned of course.