Elon Musk Says Tesla Is 'Very Close' To Level 5 Self-Driving Technology

[DrtyJrze]

I am not sure what problem this transponder would fix though. All autonomous cars need to have excellent perception that can track other cars and what they are doing on the road. So having a transponder to tell the autonomous car where other cars are and what they are doing, is not necessary. The problem is that the humans in the older cars are bad drivers. I am not sure how a transponder would fix that.

The best way to fix the problem of bad drivers is to replace the bad drivers with a capable autonomous car.

I can think of several reasons info from a “dumb“ car could be beneficial for an autonomous car:

- if turn signals are used can convey future intentions.

-if navigation is being used and destination is known, can help anticipate lane changes for exits/interchanges.

-avg speeds can assist with determining if car should overtake, or promote efficiency by drafting with the other vehicle for the portion of shared route.

- if the car was being driven continuously for long periods (4+ hours) could signal the autonomous vehicle to avoid the driver by either speeding up/slowing/changing lanes to distant itself from a fatigued driver. Or if the car is having a mechanical issue (low tire pressure) Info can be used to create a safer distance.

-if the other car hard brakes or drives “aggressively” assumptions can be made about the drivers style and the autonomous car can adjust its safety measures accordingly.

That’s just off the top of my head. Not sure how you don’t see a benefit in having more accurate information available for the car to make better decisions.

I do agree that all autonomous cars would be best, but that’s a LONG way off.

 

[mikes_fsd]

I can think of several reasons info from a “dumb“ car could be beneficial for an autonomous car:

- if turn signals are used can convey future intentions.

-if navigation is being used and destination is known, can help anticipate lane changes for exits/interchanges.

You are being very naive about how ppl drive (90% don't indicate their lane changes or turns)
Tesla (and others) can already infer/predict a "cut-in" with their NN's.

But mainly, if you do not solve for vision (that means labeling and understanding your environment on the fly) you will not be able to produce useful FSD.

 

[pilotSteve]

Dude I gotta say, reading that screen shot made me literally laugh out loud!

I’m all for the Tesla mission and I have loved my cars, but... I’m amazed this copy was approved but an entire team of people responsible for building and marketing a car like The model S. Most had to know that the 2016 model would not be able to deliver this.

.... nope, I bought FSD in 2016 with my S90D/AP2 car and expected it to be delivered as promised by 2018. When did you buyFSD?

 

[whttiger25]

.... nope, I bought FSD in 2016 with my S90D/AP2 car and expected it to be delivered as promised by 2018. When did you buyFSD?

I meant most of the engineers and developers, not customers.

i bought EAP on my 2017 model s. By the time I upgraded to my Raven last month, I knew the only EAP feature I remotely trusted was auto lane change. Summon was a gimmick that was truly useful only a couple times. So I decided not to pay the 7k. Standard AP for me now, and I’m good with it. I believe true FSD won’t be possible on this hardware, and I’ll be on my next car by the time it’s worth 7k. If Elon let me take the package with me to my next car, I would have bought it. Absent that, I’m not falling for it.

 

[whttiger25]

I think your last point is really spot on. I just did a 3,000 mile trip in my new Y, mostly Interstate highways, and it did a very good job in general. When there was construction or lots of traffic, it managed, but it did so in a way that was more tentative than a good human driver under the circumstances, which didn't mesh well with the flow. It made me think how smooth things would have been if the other vehicles had FSD systems and were communicating with each other. I do hope that we can get to FSD or close with individual vehicle sensors/software, because waiting for most cars to get it will be a long wait I suspect.

Do you or anyone know if anyone is or has developing/developed a standard NFC interface for autonomous cars to communicate with each other? Seems a standardized communication API that allows cars with different technologies and manufacturers to communicate and agree on the “state of the world” will eventually be necessary for this technology to truly transcend current limits

 

[pilotSteve]

WELL..... I have to say that HW3 is pretty darn impressive. I got my late 2016 S90D updated today to MCU2 and HW3. Wow, it truly is like a brand new car. AND....... the stop sign, traffic light, merging lanes, and general improved aptitude I experienced today on my drive home from the service center is amazing. Yes you must monitor it and approve its suggestions for traffic light and stop signs, all as described elsewhere on the forums.

BUT.... experiencing it today for the first time in my own car driving home ..... its very encouraging and maybe this will actually lead to some/most/all of what was promised as "FSD". Feels like a whole new car too!

Scottsdale service center did the update in one day as they said they would (HW3 and MCU2 upgrade on my HW2.0 2016 S90D with prepaid FSD).

My drive home was amazing! Its like getting a whole new car with these updates.

 

[goRt]

WELL..... I have to say that HW3 is pretty darn impressive. I got my late 2016 S90D updated today to MCU2 and HW3. Wow, it truly is like a brand new car. AND....... the stop sign, traffic light, merging lanes, and general improved aptitude I experienced today on my drive home from the service center is amazing. Yes you must monitor it and approve its suggestions for traffic light and stop signs, all as described elsewhere on the forums.

BUT.... experiencing it today for the first time in my own car driving home ..... its very encouraging and maybe this will actually lead to some/most/all of what was promised as "FSD". Feels like a whole new car too!

Scottsdale service center did the update in one day as they said they would (HW3 and MCU2 upgrade on my HW2.0 2016 S90D with prepaid FSD).

My drive home was amazing! Its like getting a whole new car with these updates.

That's just the increase in processing power, wait until the new software stack drops in the next few months.

 

[mrkisskiss]

It doesn't really matter if Tesla can get to something anywhere near a fully self driving car (doubtful)... being allowed to use it is a whole other ball game. As I always chime in on these threads: infrastructure changes, and V2I and V2V systems will need to be in place before any kind of autonomous driving is viable, in a meaningful way.

All this "I drive with two eyes therefore that's what autonomous cars should do" is very short sighted (both literally and figuratively speaking)

 

[goRt]

It doesn't really matter if Tesla can get to something anywhere near a fully self driving car (doubtful)... being allowed to use it is a whole other ball game. As I always chime in on these threads: infrastructure changes, and V2I and V2V systems will need to be in place before any kind of autonomous driving is viable, in a meaningful way.

All this "I drive with two eyes therefore that's what autonomous cars should do" is very short sighted (both literally and figuratively speaking)

It depends on the objective - to produce something at least as good as a human v some undefined sci-fi nirvana

 

[Battpower]

It made me think how smooth things would have been if the other vehicles had FSD systems and were communicating with each other.

Yes. But as far as I know there is no vehicle to vehicle standard communication system or protocol. Even the onboard individual vehicle parts of FSD are was off, let alone a framework for negotiation and communication vehicle to vehicle.

 

[diplomat33]

It depends on the objective - to produce something at least as good as a human v some undefined sci-fi nirvana

When you say "to produce something at least as good as a human", what is that "something"? Is it autonomous or a driver assist? That makes a big difference. Because if the goal is just a driver assist that is at least as good as a human that is a much lower bar than autonomous driving at least as good as human.

 

[mmanner]

When you say "to produce something at least as good as a human", what is that "something"? Is it autonomous or a driver assist? That makes a big difference. Because if the goal is just a driver assist that is at least as good as a human that is a much lower bar than autonomous driving at least as good as human.

I'd like full autonomous driving for some things (although I like to drive). However, it would be ok if we only ended up with a much less dangerous environment on the streets for drivers and pedestrians/bikers. We humans need some help

 

[Battpower]

Got to be very careful that your 'driver assist' doesn't lead to less attentive drivers, especially while you don't have L5.

 

[whttiger25]

It doesn't really matter if Tesla can get to something anywhere near a fully self driving car (doubtful)... being allowed to use it is a whole other ball game. As I always chime in on these threads: infrastructure changes, and V2I and V2V systems will need to be in place before any kind of autonomous driving is viable, in a meaningful way.

All this "I drive with two eyes therefore that's what autonomous cars should do" is very short sighted (both literally and figuratively speaking)

I agree with this. Because, our two eyes happen to be connected to a brain.

The AI problems I see as most difficult to solve aren't even the object identification type problems. It's the true perception problems. Anticipating a construction zone that isn't in the database, such as cones closing off a lane quickly, or road flares set up in front of an accident to divert traffic. What about difficult weather? What about avoiding road debris? What about perceiving whether or not another driver or pedestrian sees you (requiring some perception of eye contact)?

All this will be necessary for you to "feel" safe giving total control to your car, especially in a world where the vast majority of cars are still human driven. And anything less, to me at least, is a gimmick. Because I know I am a good driver. Why give my car, which is objectively a worse driver than I am, control? We are talking about life and death, or injury, as the consequence of your decision here. I see AP lane centering and adaptive cruise control as a true convenience but I still need to be fully aware of what is going on. But when it comes to handling city driving, intersections, highway interchanges, I'd rather just do it because I know I'm better than the car at it.

We can speculate about this new software stack all you want, but who is really ready to trust their cars to make left turns in front of oncoming traffic, right turns with pedestrians or across active bike lanes, etc? There is so much risk to both your life and your liability.

Don't get me started on smart summon as being somehow useful. Do you really want to have your car driving around a parking lot having no idea what damage it is causing (that you will be liable for) just so you don't have to walk to your car?

Anyway, not worth the $7k it was last month or the $8k it is now. And it won't be worth the crazy numbers Elon is talking ever to a consumer - maybe to this futuristic remote taxi fleet that is much much further away than the optimists think.

 

[DrtyJrze]

Anyway, not worth the $7k it was last month or the $8k it is now....

I’m pretty sure it IS worth $7k+, because those are the prices that people have been paying for FSD...

But hey, let’s not let a pesky little thing like facts get in the way of a good rant.

 

[whttiger25]

I’m pretty sure it IS worth $7k+, because those are the prices that people have been paying for FSD...

But hey, let’s not let a pesky little thing like facts get in the way of a good rant.

Well, it isn't worth it to me, and I was speaking for myself, and that is a fact. But my ranting is meant to throw some cold water on the claims Tesla has and continues to make in SELLING the future of the FSD package. There's lots of fine print to protect Tesla, and I highly doubt that the current feature-set is worth that much to most customers unless they have significant disposable income. Those who stretched financially to add FSD in hopes that the car will in a short time frame fully drive itself in a useful way, well, with time, we will see if it was indeed worth it to them, but I have my doubts.

On top of that, Musk threatening to keep raising the price, suckering in people to buy the package now (when it's still beta) out of FOMO, I have some ethical problems with. Most people are buying the promise, and the hope. Musk has been way behind all his promises, in the 4th dimension, for years, so I am just trying to say, buy the features now, not the hype about FSD. And if the features now are worth $8k to you, be my guest. Perhaps when it comes to cars I am still old fashioned, but I've also experienced my share of frustrations with Tesla AP tech - mainly going from AP1 to AP2 when I upgraded my Model S in 2017 and having to wait more than a year before I found AP2 as safe or useful as AP1 was, despite paying double what I paid for AP on my first car. I think I got about a $70 settlement check in the mail on that one

 

[powertoold]

99.9% of driving is less complicated than most of you guys make it out to be. Perhaps you are a great driver, super cautious and all that, but there are plenty of drivers out there with little experience or clue about how to drive. Some of them don't even know or read English.

Almost all of the "difficult" situations you guys imagine simply involve avoiding other moving objects' trajectories and stationary objects. It's not all that complicated. There's no read need for talking or communication to objects outside the car.

Sure, perhaps in super rare, once every 50k mile events, Tesla can figure out a contingency feature.

 

[diplomat33]

99.9% of driving is less complicated than most of you guys make it out to be. Perhaps you are a great driver, super cautious and all that, but there are plenty of drivers out there with little experience or clue about how to drive. Some of them don't even know or read English.

Almost all of the "difficult" situations you guys imagine simply involve avoiding other moving objects' trajectories and stationary objects. It's not all that complicated. There's no read need for talking or communication to objects outside the car.

Sure, perhaps in super rare, once every 50k mile events, Tesla can figure out a contingency feature.

You sound just like Jim Keller who said the exact same thing on the Lex Fridman podcast 5 months ago.

Sorry, not buying it. If FSD was so easy, we'd have L5 by now.

The fact is that driving might appear easy for humans, especially after we gain some driving experience. But that is because we have "solved vision" and have very advanced intelligence. We learn the traffic laws and we have brains that can problem solve very quickly. So sure, for humans seeing a situation and figuring out how not to hit other vehicles seems very intuitive. But driving for a car is far more difficult because you have to feed all the perception data into the computer that needs to interpret it, predict paths of objects, and then plan a path based on perception and driving policy. All that needs to be programmed into the computer for all cases. it does not come naturally to a computer.

I am not saying it can't be done. I'm just saying that we should not treat autonomous driving like some super easy problem because it's not.

 

[powertoold]

Tesla is solving some of the most difficult aspects of driving.

They've made amazing progress on "cut ins" based on my experience with AP. If you extrapolate their "operation vacation" strategy to other predictions, you can begin to see the indispensable value of the fleet. Tesla is able to make all manner of super-human predictions based on millions and billions of real world miles.

They've made mind blowing progress on traffic control detection country-wide. In fact, I was driving in a new city today and blew a stop sign that was blocked by a parked semi. My car knew it was there, but I didn't see it (until too late). The car is also able to recognize all sorts of light configurations and luminances in all times of day.

The fleet is absolutely indispensable for true FSD. You can begin to see it once you extrapolate Tesla's approach to FSD, based on Karpathy's talks.

I'm not saying fsd is possible. I'm just saying the fleet is indispensable on the path to reliable fsd, if it's at all possible. A lot of the human errors leading to accidents mostly involve running red lights / stop signs / bad lane changes / texting while driving / rear ends / etc. Many if not most of these errors Tesla is solving quickly, if not already.

 

[mrkisskiss]

There's no real need for talking or communication to objects outside the car

Ah - but, wouldn't you agree that even if there is no need (which is not yet proven) there is still an opportunity that's being missed here? We have, for the first time in history, an absolutely massive industry-wide investment, focused on the goal of removing the human driver whilst increasing the safety, efficiency and speed of personal transportation.

Rather than hope each individual company somehow magically solves the entire solution by emulating how humans drive, there's a rather more obvious and interesting opportunity of imbuing the entire fleet with a degree of foresight that we (as humans) would never be able to achieve...

Imagine - you could do things like optimising traffic flow based on weather conditions, automatically re-route everyone to account for major construction, even adjust each individual car's future lane choice based on relevant information - like a breakdown that's occurred further up your route, or to allow unimpeded access for emergency vehicles. Smarter infrastructure and vehicle communication would cut travel time, maximise range and increase overall efficiency whilst being safer and more widely available to all vehicle types, makes and models. Seems like a perfect fit for this current investment climate...

It seems to me like this is the obvious piece of the puzzle that's missing, and that no one seems to be doing much about.