Predictions - "Automatic driving on city streets."

[dckiwi]

With the recent update to all things FSD (packaging, pricing, description), the features coming this year have been summed up as:

• Recognize and respond to traffic lights and stop signs.
• Automatic driving on city streets.

The first one is pretty straightforward. I would expect that when autosteer and TACC are enabled, your car will pause at stop signs (and won't resume driving until other cars are out of the intersection). Same thing for traffic lights - it will slow and stop at red lights, and start again when the light turns green. I imagine the traffic light/stop sign will be rendered on the display, to demonstrate to the driver that the car sees the stop light (and to allow you to take over if it doesn't!).

However, the second bullet could mean just about anything - "automatic driving on city streets". At one extreme this could mean full L5 FSD - sleep in the car and show up at your destination. Now *of course* I don't think we're getting that this year, or any time soon. But my point is that this one sentence description really doesn't rule anything out - or in.

So my question is - what do you think "automatic driving on city streets" will mean in the next 12-24 months? Here's my completely uninformed prediction:

Imagine NoA, but on regular roads. You enter a destination and select NoA, and then engage AP. The nags will still be present. The car will be able to stay in its lane (perhaps only on marked roads, i.e. not smaller streets with two way traffic but no lines), and can change lanes to get around traffic or prepare for a turn (just like NoA will hopefully soon be able to do). It will be able to stop at traffic lights and stop signs, and will be able to make *right* turns at intersections.

When approaching your destination, the car will 'hand back over' to the driver before reaching a parking lot, driveway, parking garage, etc. It's simply not possible at this point for the car to know how to navigate/park in those scenarios.

That's my incredibly optimistic prediction. I suspect unprotected left turns won't make the first cut. I also think that it will run into issues and hand back control fairly often (similar to the 'unsupported maneuver' error that you sometimes receive from NoA). You'll still have to be incredibly vigilant at first, perhaps more so than if driving the car yourself. But man it could be fun.

What do you think? I'd rather not get into a big debate about 'Elon Time' and whether FSD is a fraud, etc. I'd love to hear some informed speculation about what people think might actually be planned by Tesla based on this incredibly broad description.

 

[pilotSteve]

that would be great!

 

[EinSV]

A conservative first step for "automatic driving on city streets" could be the equivalent of what Nav on AP is now on divided highways.

At the beginning, the system could suggest lane changes but require the driver to accept them. Once that function has enough miles to be validated on approved roads (surface streets AP currently works on or maybe a subset) the lane changes could become automatic. In other words, the same rollout as Nav on AP. Right hand turns might come later (and then left turns) -- in the meantime the system could notify the driver that a turn is coming and ask him/her to take over to make the turn.

I use automatic lane changes on surface streets often already and it seems to work well for me. This seems like a natural progression.

 

[aronth5]

I was surprised they didn't say secondary roads as an easier way to introduce FSD incrementally. I'm sure many already use EAP for secondary roads and it does pretty well. Just add stop sign and street lights and state advanced support for things like rotaries and city streets will be added over time. But by stating city streets do they really mean I should try FSD in Boston? Not a chance!

 

[OPRCE]

My list, in order of priority, of essential features for release in CityAP:

1. stopping for pedestrians in or moving into the planned path, whether jaywalkers or those at a zebra-crossing without lights.

2. stopping for crossing bicycles, otherwise shifting over in lane to leave ample space.

3. stopping reliably for stationary vehicles, also on corners.

4. stopping at red lights/stop signs.

 

[Troy]

"automatic driving on city streets". At one extreme this could mean full L5 FSD - sleep in the car and show up at your destination.

Hardware 3.0 won't have this level of autonomy based on Elon in the podcast here. It will require human oversight. What you will get is Navigate on Autopilot in the city with traffic light and stop sign recognition and auto lane change without confirmation if you have FSD.

Tesla was originally planning to achieve driverless Autopilot with HW2.5. The idea was that your car would provide driverless rides while you are at work. It would either provide rides to friends and family or you would be able to add it to the Tesla Network. Elon even said you would be able to select whether only 5-star customers should be able to book rides. However, that plan is now ditched until HW4.0.

 

[EinSV]

Hardware 3.0 won't have this level of autonomy based on Elon in the podcast here. It will require human oversight. What you will get is Navigate on Autopilot in the city with traffic light and stop sign recognition and auto lane change without confirmation if you have FSD.

Tesla was originally planning to achieve driverless Autopilot with HW2.5. The idea was that your car would provide driverless rides while you are at work. It would either provide rides to friends and family or you would be able to add it to the Tesla Network. Elon even said you would be able to select whether only 5-star customers should be able to book rides. However, that plan is now ditched until HW4.0.

That's completely false.

In the Ark interview, Elon predicted HW3 would be "feature complete" by 2019 (driver oversight still required) but by the end of 2020 drivers could sleep in their cars. The timing of the latter prediction is certainly debatable but in no way, shape or form did Elon suggest HW3 would not achieve autonomous driving. Here is Ark's summary:

"Elon told us that he is certain that Autopilot will be “feature-complete” for self-driving by the end of this year and he believes that, by year-end 2020, people will be able to relinquish all responsibility to Autopilot and fall asleep while being transported from point to point, pending regulatory approval." Takeaways from Our Recent Podcast with Elon Musk
​

Elon's quote:

“My guess as to when we would think it is safe for somebody to essentially fall asleep and wake up at their destination? Probably towards the end of next year. That is when I think it would be safe enough for that.” Elon Musk: Full Self-Driving Teslas This Year, "Unequivocal" Tesla Autopilot Improves Safety | CleanTechnica
​

 

[Sonic_78]

These features are rolling out differently than I had anticipated. Originally, I had considered Tesla would first roll out the stop sign, stop light detection as a safety feature. Along the lines of emergency braking, lane keeping, etc... This way Tesla could begin to test the features but under the guise of a driver safety feature.

If Tesla could alert the driver they’re about to run a stop light / stop sign that alone would be a great step forward in driver safety.

 

[aronth5]

Hardware 3 is not going to address inclement weather like snow and ice that covers up cameras. There are numerous posts where owners received messages that EAP was disabled due to visibility problems/inclement weather. So 4.0 hardware will come since hardware always improves but 4.0 will also need to address when visibility problems disable EAP/FSD.

 

[electronblue]

That's completely false.

In the Ark interview, Elon predicted HW3 would be "feature complete" by 2019 (driver oversight still required) but by the end of 2020 drivers could sleep in their cars. The timing of the latter prediction is certainly debatable but in no way, shape or form did Elon suggest HW3 would not achieve autonomous driving. Here is Ark's summary:

"Elon told us that he is certain that Autopilot will be “feature-complete” for self-driving by the end of this year and he believes that, by year-end 2020, people will be able to relinquish all responsibility to Autopilot and fall asleep while being transported from point to point, pending regulatory approval." Takeaways from Our Recent Podcast with Elon Musk
​

Elon's quote:

“My guess as to when we would think it is safe for somebody to essentially fall asleep and wake up at their destination? Probably towards the end of next year. That is when I think it would be safe enough for that.” Elon Musk: Full Self-Driving Teslas This Year, "Unequivocal" Tesla Autopilot Improves Safety | CleanTechnica
​

I think what many people are saying is that after all the previous hyperbole from Elon, they don’t believe him on this one anymore. His past predictions on autonomy do not really support such belief.

I would have to agree with those sentiments. I find the suggestion that people can safely sleep in the driver’s seat of their Tesla at the end of 2020 to be impossible — at least assuming Tesla sticks to camera only/mainly and does not introduce some other suite like 360 degree Lidar or such.

There is this nagging feeling that’s hard to shake. For all his impressive achievements, Elon may have gotten autonomous driving terribly wrong.

 

[diplomat33]

My guess is that "automatic city driving" will be navigate on autopilot in the city. You set a destination, engage AP, you get the one blue line in the center and the car will stay in the lane, change lanes and also make turns at intersections based on your navigation directions.

 

[aronth5]

My guess is that "automatic city driving" will be navigate on autopilot in the city. You set a destination, engage AP, you get the one blue line in the center and the car will stay in the lane, change lanes and also make turns at intersections based on your navigation directions.

Secondary roads sure, but driving in downtime Boston in rush hour with pedestrian and bikes being as unpredictable as a running rabbit is not something I will try anytime soon. However on secondary roads FSD should work very well and I'm very excited about trying that. My point is just that full FSD is a long journey and referring to city driving as the next phase instead of secondary roads is not great judgement and will come back to bite Tesla.

 

[Thp3]

I hope that “automatic city driving” means that autopilot won’t shutoff when going through an intersection, a place without painted lanes. Where I live, with unpainted lanes, the autopilot routinely gives up when going through an intersection.

I think city driving will require more input from the gps and maps than the current system has.

I’m looking forward to the improvements. I’m sure that even with “Elon Time”, it will happen far faster than my streets getting good visible painted lanes.

 

[Soda Popinski]

I'm hoping / guessing "automatic driving" means something like EAP but on city streets. It'll stop at lights, stop signs, not run over pedestrians, take turns at intersections, etc, but still require the driver to keep a hand on the wheel. We may not get all of that until "feature complete" so maybe a subset of those functions. I'd guess turning might come in last.

 

[EricOtown]

I think the 2nd bullet essentially means that all of the autopilot features that currently work from on the highway from on ramp to off ramp will work off the highway on most streets. This would include auto-steer, auto lane change, and navigate on autopilot, as well as the annoying and frequent alerts asking you to make small movements of the steering wheel to ensure that you are still wake and alert.

What remains to be scene is if this new functionality will work on all non-highway streets or just main streets with well defined solid and dashed white lines defining the sides of the street and the dividers between lanes.

For example, the roads in and around my neighbor do not have any white line markings to define the sides of the road or to divider lanes of traffic moving in opposite directions.

But once I get out of my neighborhood, there are major divided roads with solid white lines defining the sides of the road and dashed lines, concrete dividers, and landscape dividers between opposing lanes of traffic.

Will these autopilot features work on all roads or just the latter. That’s the big question.

 

[SalisburySam]

I like the interpretations above, and I hope that at least some of them are included in the broad phrase “automatic driving on city streets.” That said, I doubt it for anytime soon, and would be very surprised if any of that happened in the next 3 months. While supposedly amazing things are happening in software under NDA, the delivered stuff remains far behind. My v32.2.2 does little that the last several versions didn’t already do to some extent, and nothing that helps me in any substantial way. Still can’t use Basic (Stupid, Backwards, Elementary, Original, Whatever) Summon, AutoPark still never shows even when trying the techniques in the fora where others have made it work, and the phantom braking still happens such that I can’t use TACC and the rest of the auto-stuff with passengers in the car. And this is on a 14-month old vehicle with EAP/FSD since birth. My expectations are tempered by my experience with existing software so forecasting functionality to be included in the near future is subject to a lot of skepticism from me. I love what I’ve got, I just have very low expectations for anything approaching FSD Levels 4 or 5, and only slightly higher thoughts about Level 3. Sorry to be a Dougie Downer, but I just don’t see it, and I pretty much disregard promises that I can’t see IRL. I would love to be proven wrong.

 

[voyager]

I guess we all are waiting for Elon Musk's prospect that Teslas can be fully (Level 5) Autonomous in 2020 will become reality.
Other developers seem further behind.

My general guess is 'fully' autonomous outside the city, manual mode within city limits.
A warning light flashes on when you pass outer markers.
And not some grey zone in between, with different self-driving modes.

Interesting read on how to fool a deep-learning system.
Your Car May Not Know When to Stop — Adversarial Attacks Against Autonomous Vehicles

 

[Daniel in SD]

My prediction is that “automatic driving”
on city streets“ will never be released as a system that requires driver monitoring (SAE level 2). I think we will see some sort of emergency braking for stop lights and stop signs.

 

[diplomat33]

My prediction is that “automatic driving”
on city streets“ will never be released as a system that requires driver monitoring (SAE level 2). I think we will see some sort of emergency braking for stop lights and stop signs.

It is entirely possible that Tesla will wait until "automatic city driving" is L4 before releasing it. And Tesla could deploy "automatic city driving" in shadow mode at no risk and use that to validate when it is safe to release as L4.

 

[electronblue]

It is entirely possible that Tesla will wait until "automatic city driving" is L4 before releasing it. And Tesla could deploy "automatic city driving" in shadow mode at no risk and use that to validate when it is safe to release as L4.

I don’t see how that could be reconciled with the fact that Tesla is saying in Design Studio to expect Automatic city driving later this year (I’m not sure shadow mode would count...). Unless you don’t expect it out this year?

While we know Tesla is on record about Level 5 no geofence feature complete in 2019, the autonomous release was said to be in 2020. Level 4 requires autonomous driving so that coming out in 2019 seems even iffier than usual.