Elon: "Feature complete for full self driving this year"

[metaxaos]

A year is not so long. I'd be willing to consider 5 years to be "soon." Of course, others of Musk's promises have specified dates that have passed. But the original robotaxi promise specified no date. But as soon as the cars first sold with the promise of FSD start to wear out from normal use, if they can not yet drive themselves without a human in the car, Musk has broken his promise.

I don't think he can accomplish this on these cars because I think more sensors are needed. Sensors are the reason why I bought EAP but not FSD. I'll buy the FSD car when it becomes available. And I'll buy it as soon as I can.

Well, personally, I don’t care about robotaxi at all. I’m neither giving up my personal car to it, nor sharing my car as robotaxi unit, even if it’s available. I don’t demand level 5.
What I know (given early autonomy day feedback) is that Tesla must’ve achieved at least conditional level 2 city autopilot which is not perfect, yes, requires constant attention, yes, but it exists, and given enough time passed since that demo, should be able to drive you through typical city scenarios while supervised. And I want it under my responsibility.
You know, highway Autopilot is not prefect. I faced at least a couple of situations when it was clearly about to swerve me into divider, so I had to intervene. And yet it’s released. So I believe it should be possible for city NoA as well. The key point here is to make proper indication about car’s upcoming decisions, so that driver has enough time to react. That would be enough maybe not for everyone, but for most of those who constantly monitor any updates and nags about delays. Give them something! Perfect solution might be deployed later.

 

[Barklikeadog]

A year is not so long. I'd be willing to consider 5 years to be "soon." Of course, others of Musk's promises have specified dates that have passed. But the original robotaxi promise specified no date. But as soon as the cars first sold with the promise of FSD start to wear out from normal use, if they can not yet drive themselves without a human in the car, Musk has broken his promise.

I don't think he can accomplish this on these cars because I think more sensors are needed. Sensors are the reason why I bought EAP but not FSD. I'll buy the FSD car when it becomes available. And I'll buy it as soon as I can.

"I feel very confident predicting 1 million autonomous robo-taxis for Tesla next year," "Not in all jurisdictions because we won't have regulatory approval everywhere, but I'm confident we will have regulatory approval at least somewhere, literally next year. " "If you fast forward a year, maybe a year three months, we'll have over a million robo-taxis on the road."

 

[dckiwi]

You can count me as one of those. I can recite edge-case scenarios all afternoon that I doubt Tesla will solve in my lifetime...

* snow covered road obscures all lane markings
* newly paved interstate without lane markers
* policeman standing in the middle of the road directing traffic around an accident
* '67 VW rear bumper tastefully decorating your lane
* ladder flies out of the pickup truck in front of you -- which way will it bounce?
* intersection changed from stop sign to roundabout since the last time maps were updated
* there's a 2 foot deep puddle in your way
* the car misses the "20 MPH SCHOOL ZONE" sign

Some of these are fair but some are ridiculous:

"ladder flies out of the pickup truck in front of you -- which way will it bounce?"

Tell me Neo - could *you* predict in a split second which way a falling ladder will bounce? lol

The reality is that most human drivers, when they encounter an unexpected scenario at high speed, just slam on the brakes and swerve in an arbitrary direction.

 

[dckiwi]

First let me say that I agree with everyone's skepticism about the likelihood that City NoA is around the corner.

However there is one (unlikely) possibility to consider - Tesla has City NoA working well on HW3, but they're holding off until the retrofits are well under way so that the HW 2.0/2.5 owners don't bust out the pitchforks when it rolls out to new owners. They are clearly just getting started with the 2.5 retrofits in California.

 

[Barklikeadog]

Some of these are fair but some are ridiculous:

"ladder flies out of the pickup truck in front of you -- which way will it bounce?"

Tell me Neo - could *you* predict in a split second which way a falling ladder will bounce? lol

The reality is that most human drivers, when they encounter an unexpected scenario at high speed, just slam on the brakes and swerve in an arbitrary direction.

He suggested that 'ridiculous' situation because it apparently happened to him.

I saw a tire flopping on a semi, immediately got in the other lane and slowed down. THWAP! Rubber shards shoot out, would have gone through my windshield had I not been paying attention and moved.

 

[evp]

Some of these are fair but some are ridiculous:

"ladder flies out of the pickup truck in front of you -- which way will it bounce?"

Tell me Neo - could *you* predict in a split second which way a falling ladder will bounce? lol

The reality is that most human drivers, when they encounter an unexpected scenario at high speed, just slam on the brakes and swerve in an arbitrary direction.

In this case, I had a few seconds watching it move around unsecured in the pickup before a big bump tossed it out. I took that time to determine my escape plan if it came out. So I could react while it was still in the air, swerving into the lane I already knew was open. I pity the people behind me on the interstate who had to deal with a ladder suddenly blocking their lane.

Of course if it had landed feet first instead of flat, nobody could predict what it would do next.

 

[conman]

Tesla has City NoA working well on HW3, but they're holding off until the retrofits are well under way so that the HW 2.0/2.5 owners don't bust out the pitchforks when it rolls out to new owners.

Sounds possible, but we can't bust out what we're already waving around.

 

[lambodiablo205]

Sounds possible, but we can't bust out what we're already waving around.

I believe it's a mix of 3 things...

1) They can't release it until it's super reliable. They know people abuse the system, so you can imagine how bad it will be when people just let the car handle every light and stop sign without supervision. All it takes is 1 car to run through a red light and cause a death. (This has actually already happened - A tesla ran a red light in San Francisco killing another person, and everyone blamed autopilot because it didn't stop at the light, even though autopilot doesn't react to lights).

2) MONEY! Q1 is usually a pretty poor quarter due to lower sales (seasonality), so with them releasing it in Q1 instead of Q4, they are able to recognize all that revenue from Full Self Driving purchases over the last several years. They will have until the end of March to release it and recognize that revenue.

3) Retrofits - people will be pissed if full self driving is available, but their car doesn't have all the newest features.

 

[dckiwi]

In this case, I had a few seconds watching it move around unsecured in the pickup before a big bump tossed it out. I took that time to determine my escape plan if it came out. So I could react while it was still in the air, swerving into the lane I already knew was open. I pity the people behind me on the interstate who had to deal with a ladder suddenly blocking their lane.

Of course if it had landed feet first instead of flat, nobody could predict what it would do next.

That's an interesting, albeit incredibly rare, situation. Here's a more common one. Last week I was rear-ended in my Model 3 while sitting at an intersection. The woman behind me got distracted - most likely looking at her phone. This is the third time in my life I've been hit by a distracted driver. If she had AP this wouldn't have happened.

I'm just saying, if we're going to talk about how incredible human drivers are in falling ladder situations, we should also talk about how woefully inadequate they can be with even more basic driving tasks.

 

[jeewee3000]

That's an interesting, albeit incredibly rare, situation. Here's a more common one. Last week I was rear-ended in my Model 3 while sitting at an intersection. The woman behind me got distracted - most likely looking at her phone. This is the third time in my life I've been hit by a distracted driver. If she had AP this wouldn't have happened.

I'm just saying, if we're going to talk about how incredible human drivers are in falling ladder situations, we should also talk about how woefully inadequate they can be with even more basic driving tasks.

Fully agree.

I commute daily since 2011, about 30 to 45 minutes either direction.

At least once a week my commute is slowed due to a traffic accident, which I have to pass by if Google Maps couldn't warn me ahead of time.

These accidents are largely (+95%) due to a clear error on behalf of one of the involved drivers. Most common errors are:
- lack of control of the vehicle (including speeding and therefore not having enough reaction time when their driving space gets blocked);
- lack of attention to the road (staring at mobile phones, GPS, radio and even DVD-player or tablet (happens a lot with semi drivers))
- lack of communication with other drivers regarding your intended actions (i.e. not using indicators, sudden acceleration/deceleration or turns, ...)
- lack of defensive driving (harder for FSD to solve IMO, but some people just don't seem to grasp what situations might become dangerous in a few seconds).

Basic FSD (feature complete) would at least combat the first three of the above points, rendering it very safe IMO. We'll just have to wait and see how effective it can be regarding edge cases in the near future. (H1 2020)

 

[diplomat33]

Elon's definition of "feature complete":

Yeah, feature-complete, I mean, it’s the car able to drive from one’s house to work, most likely without interventions. So it will still be supervised, but it will be able to drive...So feature-complete means it’s most likely able to do that without intervention, without human intervention, but it would still be supervised.

Safety is a huge part of it of course. You can't release traffic light response if the car is going to run through red lights. But note that Elon is promising driving from your house to work with almost no interventions. That's easier said than done. Tesla is probably encountering all kinds of cases that they had not thought about before that they need to solve in order to reduce the interventions down to Elon's standard of "most likely without interventions". That's probably a big reason why Tesla has not released "feature complete" yet. Tesla probably has some version of "feature complete" running on AP3 but it probably has too many interventions so it is not good enough to release yet.

 

[banned-66611]

All Tesla Cars Being Produced Now Have Full Self-Driving Hardware

Obvious lie. They don't have any way to clean the cameras, for example, so if you summon your car from the other coast and it gets splashed with a bit of mud... Well what happens, does it stop in the middle of the highway like autopilot does?

Their hardware is far, far away from what they promised for FSD.

 

[daniel]

Well, personally, I don’t care about robotaxi at all. I’m neither giving up my personal car to it, nor sharing my car as robotaxi unit...

I would never use my car as a robotaxi. However, the term is useful, amid all the confusion about what FSD is supposed to mean. Ultimately, I would like to be able to sleep in the back while my car takes me where I want to go. That's "robotaxi-capable" even if I'm the only one ever using my car.

I don't expect this within the next ten years, and at my age I might not live to see it at all.

"I feel very confident predicting 1 million autonomous robo-taxis for Tesla next year," "Not in all jurisdictions because we won't have regulatory approval everywhere, but I'm confident we will have regulatory approval at least somewhere, literally next year. " "If you fast forward a year, maybe a year three months, we'll have over a million robo-taxis on the road."

Yep. That's the over-promising.

Regarding the idea that Tesla may have an adequate City NoA but are holding it back until all the older cars with the FSD package have gotten the HW3 computer: I doubt this is true, but if it is, it once again highlights what a terrible idea it was to sell something that didn't yet exist. Musk confidently stated that the cars had all the hardware needed for FSD, before FSD existed. Since FSD didn't exist, it was not possible to know what hardware would be needed.

Musk is a brilliant businessman, a great visionary, and it's his optimism that gave us these cars in the first place. But it was just plain stupid to assert what hardware would be needed for a task that had not yet been implemented. And the result of that stupid decision is that the first people to pay for the feature may be the last to get it, and may very well in the end be left out entirely when (as I expect) it turns out that even more hardware is needed, which cannot be retrofitted into these cars. And at the time it currently takes to retrofit a car, it will take them ten years to do them all. By which time HW3 will be obsolete.

 

[Soda Popinski]

In this case, I had a few seconds watching it move around unsecured in the pickup before a big bump tossed it out. I took that time to determine my escape plan if it came out. So I could react while it was still in the air, swerving into the lane I already knew was open. I pity the people behind me on the interstate who had to deal with a ladder suddenly blocking their lane.

Of course if it had landed feet first instead of flat, nobody could predict what it would do next.

This begs the question, what would you do if you were the driver behind instead, with a ladder suddenly blocking their lane?

 

[DigAdrenaline]

First let me say that I agree with everyone's skepticism about the likelihood that City NoA is around the corner.

It's already rolling out. Sort of.

The way this will play out is like highway NoA. You'll teach the car to stay in it's lane. Separately you'll add that if I tap the turn indicator, it will automatically change lanes. Separately, it'll keep speed with traffic both via radar and maps. Separately, it'll do other things. Once all of them are 99.99% (or whatever) reliable, you glue them together and THAT is highway NoA. All highway NoA is, is the glue connecting them all together, and that glue might mean the 99.99% individually regresses back to 99%. So we see some regression but march forward again. After the fact, we'll add traffic pylons and emergency vehicles and whatever else.

City NoA will be the same. We'll see stop signs. Traffic Circles. Lane Keeping. Pedestrians. Cyclists. Intersections. Smart Summon. Smart Park. etc. etc.. etc... Then, once they all mostly work individually, it will all be glued together. It would be stupid to try to drop one big pile of software when there can be incremental upgrades over time that get us there. It's much more likely to succeed, and it shows progress along the way.

Some parts of the puzzle are already built. Some were built for highway NoA and continue to improve there for both situations. Others are new and many are still yet to come. But seeing the display for stop signs and traffic lights is a first step. It used to be hidden, but now it isn't. The IC doesn't need to display what it sees in order for it to act on it, but it builds confidence in the system and heightens the confidence and ability of the driver to interact with the computer as it drives, so although technically un-necessary, it's a step forward that gets us closer to self-drive.

 

[DigAdrenaline]

They are clearly just getting started with the 2.5 retrofits in California.

They are? What did I miss? Can you point at a link for that?

 

[diplomat33]

Investors will be asking 2 questions related to feature complete and FSD at the earnings call later this evening.

1. You set expectations that you would be feature complete on FSD by the end of 2019. Can you please provide an update on when we may see this with end users? Where are you in retrofitting the FSD computer to older models?

2. Elon, can you give us an update on the Dojo project you talked about at Autonomy Day? Beyond that, any details on Tesla’s self-supervised learning efforts—along with how these efforts are helping Tesla achieve FSD capabilities—would be great.

Check back in this thread, and I will post Elon's answers.

 

[am_dmd]

I wish someone asks specifically AP2/MCU1 as an older hardware regarding retrofit. Let's see what BS answer we get this time

 

[selfbp]

I wish someone asks specifically AP2/MCU1 as an older hardware regarding retrofit. Let's see what BS answer we get this time

This should be question number 1 for technicalities of Tesla this call. Thousands of us are in the dark and at this point I just want an answer good or bad.

 

[mongo]

I wish someone asks specifically AP2/MCU1 as an older hardware regarding retrofit. Let's see what BS answer we get this time

Just checking that you are aware of this Dec 19th Tweet.

This is quite a thorny hardware problem, as there are many diff versions of mcu & autopilot computer & supporting hardware. Cars last so much longer than phones! Hopefully able to upgrade mcu1 & ap2.0 in a few months.