Traffic lights, stop signs, and city driving

Traffic lights, stop signs, and city driving

So the Tesla website still states under FSD for custom orders:

“Coming later this year:
Recognize and respond to traffic lights and stop signs.
Automatic driving on city streets.”

When I first started seriously looking at Tesla a few months ago I thought “sure, later this year...ok”

When I placed my order almost 2 months ago and the statement was still there I thought “I bet it’ll be changed to delayed till early 2020 or something pretty soon”

Now after having the X for a month and it being just two months from the end of “this year” the site still has the statement.

Will FSD really recognize and respond to traffic lights & stop signs and drive in the city by the end of the year???

That would be epic! What do you think?

jimglas | 31 octobre 2019

nobody knows

bp | 31 octobre 2019

It's likely this capability has been working for a while with Tesla's internal test vehicles - and that Musk's car has had this capability.

Because of the lack of standardization with traffic lights & signs (including speed limits), it would not be surprising for it to take longer than Tesla has projected to get this reliably working - and reliably read traffic lights & signs.

While we might get this feature in the next few months - would not be surprised for this to slide into next year - and even when it does get released, also wouldn't be surprised that the software will miss or misinterpret some signs until Tesla uses fleet learning to mature the software to the point where this capability is stable enough to enable FSD.

sunsinstudios | 31 octobre 2019

@bp I bet you’re right that it may get released with the caveat that it is beta and therefore may miss some signs. I could live with that.

I think Tesla is doing an amazing job and I just don’t want to take a hit (from the press, disatisfied owners, haters, Porsche owners lol) for being misleading since it states “this year”. I would even be fine with “eventually”.

@jimglas who knows it may happen after the price goes up in November! =] would be dope!

Saxman | 1 novembre 2019

How do you suppose FSD will deal with 4-way stop sign intersections? Theoretically its, First Come First Served. How will it "know" when it is its turn to proceed at busy intersections? | 1 novembre 2019

@Saxman - Probably just like humans - you wait for another car to go, and then cautiously move into the intersection. Waymo has been doing it for years in our neighborhood. I usually see at least one Waymo car every day I'm out and about. Lots of 4-way stops in our area too. I've been behind one at a 4-way stop at least 4-5 times now just this year. It seems to figure it out just fine.

Saxman | 1 novembre 2019

Great to know.

Thanks TeslaTap.

Technistore | 2 novembre 2019

It will just turn off the cameras and go for it.

Technistore | 2 novembre 2019

It will just turn off the cameras and go for it.

Technistore | 2 novembre 2019

It will just turn off the cameras and go for it.

Technistore | 2 novembre 2019

It will just turn off the cameras and go for it.

Technistore | 2 novembre 2019


sunsinstudios | 3 novembre 2019

I think any driving conditions that humans can navigate, a computer can learn to navigate. For the four way stop, the vehicle can be programmed to wait for the car on the left to go before proceeding. At round-a-bouts it will probably just go slower (or speed limit lol) and merge as needed while sensing vehicles. There are decision making algorithms that we all follow in these cases; it’s second nature so it seems like instinct after many years of driving. Weather like fog may be tricky but perhaps radar is superior in that scenario.

Right now I notice the car wants me to take over immediately when one lane turns into multiple suddenly such as after an off ramp but even “tricky” lane marking can be handled with rules such as “stick to the right or left lane marking”.

Overall, excited for the future in this car!

nukequazar | 3 novembre 2019

Waymo has 3D HD maps of the areas they drive in, and LIDAR on the cars. It will be interesting to see how close we get to FSD with the sensors we have.

bp | 4 novembre 2019

Evidently Tesla was able to book part of the FSD funds when they released the Smart Summon feature [though doing that is questionable - since that feature is clearly early beta - and not close to be an "official" version - which should be the trigger for booking the FSD funds].

If Tesla is able to book more of the FSD funds when they release street/traffic sign/light detection, that could provide Tesla incentive to release that functionality earlier, even if it is in "early beta" state.

3D HD maps work great - in a static environment, when nothing changes.

But that's not the real world - and when the 3D HD maps don't match current conditions, then the onboard self-driving systems must adapt to the current environment - construction barriers, changed/temporary lane markers, emergency responders, potholes, …

So, if the onboard systems must be able to understand current conditions - then all self driving vehicles will need to be able to do that without relying on 3D HD maps - and use onboard sensors (cameras, radar, lidar, proximity sensors, …) along with object detection capability to figure things out.

In theory, since humans are able to drive with the equivalent of two cameras, with limited visibility (head movement, mirrors) - and Tesla's AP systems have continuous 360 degree camera coverage, proximity sensors for blind spots in the immediate vicinity of the vehicle and forward facing radar - Tesla (Musk) may be right in claiming they have sufficient sensors - as long as they have enough processing power to interpret the data and determine how to use it. | 4 novembre 2019

Elon also stated in Automony Day talk they went down the HD mapping approach and discovered it really was a bad way to go. They didn't elaborate much though. I expect HD mapping allows for a sloppy autonomous driving system, where FSD really required understanding the current environment in fine detail.

Lidar is cool but so expensive and unsuitable for a production car. It also requires massive additional compute power to turn the data into something useful. I'm sure someday, it might be viable on a $200K car, but it's still a long way off from being production-ready. It does make for a quick shortcut to getting FSD working in demo cars, when cost and mainenance are not considerations.

nukequazar | 5 novembre 2019

@TT, those two things--3D HD mapping and lidar--are what enable those Waymo cars to drive autonomously around your neighborhood, as you mentioned. They are not driving autonomously with Google Maps and the sensors we have. | 5 novembre 2019

Agreed. Too bad it may take 10+ years for the Google solution to be affordable. It's a quick way to make demo cars work though, and I'm sure they learn a lot from that.

Pungoteague_Dave | 5 novembre 2019

@sunsinstudios "I think any driving conditions that humans can navigate, a computer can learn to navigate. "

But it needs the eyes and ears to do it. How will Tesla ever deal with emergency vehicles, police officers manually directing traffic by hand, etc.? When we hear a siren, we look around and find it. The Tesla cameras on our AP3 cars have limited range and focus, and without a microphone, would be clueless to look for the emergency vehicle anyway - even if it could quickly pull out of the way in an intelligent manner. Not likely ever on our cars, or anytime soon on any Tesla - see the Smart Summon videos out there. Mine ran into a boat trailer (heavy steel frame) yesterday to the amusement of my onlooking guests. FSD is hard. And a looooong way off

sunsinstudios | 6 novembre 2019

@Pungoteague_Dave sorry for your accident, that sucks, hope the damage was not extensive.

I think there will be niche scenarios where the driver will have to intervene such as in cases of heavy construction, officers directing traffic, flooded roads, downed powerlines, plywood flying off trucks, and general post-apocalyptic scenarios.

For emergency vehicles I think the cameras could be programmed to detect blue and red flashing lights and move over safely; like deaf people must do. Not impossible.

10 years ago, when the roadster came out, no major manufacturers were building EVs, now most (if not all) major manufacturers have at least 1 concept car in their fleet. 5 years ago AP came out, now we are debating FSD. Nissan has propilot and other manufacturers will have to follow. To me, 95% FSD (oxymoron?) can be a reality in two years.

"Looooong way off" is sooner than we think | 6 novembre 2019

Tesla has two microphones in every car. There is no reason they couldn't listen for sirens and other activities. Even though the microphones are inside the car, it's fairly easy to filter out any playing audio to be even more sensitive than humans.