Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD Beta 10.69

This site may earn commission on affiliate links.
Not sure what you think those pics are for? Anything less than 4 wheels is a motorcycle- legally- in the US anyway.
Actually, the definition of “motorcycle” varies from state to state. For example, in Nevada, the DMV calls 3-wheelers “trimobiles” and they do not require a motorcycle license, and you’re not required to wear a helmet with some configurations.


So you may think that they’re motorcycles, but I’ll go with the definition and codes applied by the DMV of whatever state I’m in.
 
  • Like
Reactions: sleepydoc
Could it be a camera calibration issue on the left?
It was still in front, so would still be using the same camera.

Also just because it is not display doesn't mean it will hit it.
I know that, but if you read the original post, I had to hit the brake to keep it from hitting the vehicle. I waitied until it was real close because we were going slow. FYI, it had the single wheel int he back, and was in-closed, which is why I called it a car, since it looked more like a car than a motorcycle. Though not sure why it matters to you what I called it, since the description was enough to know what I was talking about.
 
Actually, the definition of “motorcycle” varies from state to state.

ACTUALLY- Funny thing about US law-- federal has supremacy over state.



actual federal NHTSA definition of a motorcycle said:
Motorcycle means a motor vehicle with motive power having a seat or saddle for the use of the rider and designed to travel on not more than three wheels in contact with the ground

All the vehicles in those pictures, under federal law in the US, are motorcycles- their manufacturers are required to meet the federal safety standards of motorcycles-- which are much less than for cars.

States are of course free to impose their own sets of rules and definitions regarding standards for operating or registering any given vehicle config within their state-- but that doesn't change the federal definition of the vehicle or which set of federal safety standards it must comply with.
 
I haven’t read all 208 pages of this thread so excuse me if this is an issue that has been discussed, but the speed limit in my neighborhood is wrong.
My block is windy, narrow, and without paint nor signs, but the car thinks It’s a 35 MPH road. When on an adjacent, similar, road the limit is 25. As soon as I turn into my street (which is only ¼ mile long) it jumps to 35 The same limit shows in my driveway … shouldn’t limit disappear when off road…or in my garage???
Other times when on a four lane wide painted road it defaults to 25 MPH (until car passes the 45 limit sign).
I assume It’s a mapping error and not an FSD glitch, but thought I’d ask if anyone experienced similar issues. I can’t say for sure but I don’t think this problem was evident in earlier versions of FSD.
Is there a way to correct speed limit problems?
I tried emailing the FSDbeta mailbox and posted this Thanks!
 
Um. One Day, the town I live in decided that All Roads were going to have a speed limit of 25 mph. It took quite a time for the signage to catch up. And even longer for the GPSs to do so: but the all eventually got there, sign or no sign.

One can see that NAV sort of knows what the speed limit is on a road it turns onto, because if one dials up a higher speed, when the first speed sign shows up, at the original speed limit.

Therefore, I suspect your issue is with the map data on the car. That gets updated periodically. From the copyright notices, that data is sourced mainly from Google. There’s rumors that Tesla updates that. Perhaps by crowdsourcing.

Don’t know how to get a quick fix, but hitting the right scroll wheel each time this occurs and intoning the words, “Bug report wrong speed limit should be 25 not 35” might help.
 
  • Like
Reactions: sleepydoc
ACTUALLY- Funny thing about US law-- federal has supremacy over state.





All the vehicles in those pictures, under federal law in the US, are motorcycles- their manufacturers are required to meet the federal safety standards of motorcycles-- which are much less than for cars.

States are of course free to impose their own sets of rules and definitions regarding standards for operating or registering any given vehicle config within their state-- but that doesn't change the federal definition of the vehicle or which set of federal safety standards it must comply with.
Interesting how far this has veered off from the original point of how FSD renders 3-wheelers. But whatever…

So your definition of “motorcycle” is based on the Federal description, which defines what safety standards the vehicle is required to meet Based on the number of wheels it has.

Mine is based on the definitions of the state you’re in, which defines the type of operating license and safety equipment the user is required to wear.

I’d argue the state definition is more relevant, since it’s the one your local DMV will go by when you try to license, register, and operate your vehicle.
 
Speed limits are a long standing issue with no clear answer. It’s not clear where Tesla gets its data from, I’ve seen several people say TomTom and others say Open Street Maps.org. There’s a road near my cabin that is 55 MPH but Tesla thinks it’s 35. I found out the data on TomTom said 35 so I submitted a correction that was finally accepted a month or so ago. Still waiting to see if it changes in Tesla’s Nav system and how long it takes.

The other problem is what speed should Tesla use when it doesn’t have any information or if there‘s a discrepancy between the database and the camera. It’s easy to say “obviously it should use the camera data,” but what about times where the sign is obscured by another car? In reality, humans use a combination of personal recall, actual signage and other visual cues. How do you program that?
 
....and THE ENTIRE point is Tesla doesn't render autocycles properly. I'm just trying to explain why that is likely and get dragged down this rabbit hole. seeepydoc why don't you contact Tesla and TELL them what you think a car is so they can render it since you know what it should be.
Well, you were the one who claimed there were no 3 wheeled cars and then brought up an obscure bureaucratic definition that only applies to 2 countries in the world to explain why it would matter to the FSD computer.

edit: about 10 years ago I read a story of a man who fashioned a motorized recliner and got a DUI on it. Evidently a Lay-z-boy is a car, too!

I posted my hypothesis above but personally, I really don’t care if it renders 3 wheeled vehicles as cars or motorcycles or rickshaws as long as it doesn’t run into them. If the problem Stems from the system not being able to categorize the object then it’s a poor design choice. As a human, if I see something I can’t identify, I still avoid it. It sounds like Tesla may finally be starting to do that with unidentified obstacles.
 
Last edited:
Interesting how far this has veered off from the original point of how FSD renders 3-wheelers. But whatever…

So your definition of “motorcycle” is based on the Federal description, which defines what safety standards the vehicle is required to meet Based on the number of wheels it has.

Mine is based on the definitions of the state you’re in, which defines the type of operating license and safety equipment the user is required to wear.

I’d argue the state definition is more relevant, since it’s the one your local DMV will go by when you try to license, register, and operate your vehicle.

I'd argue the opposite-- because if you want an AI to correctly classify what it is seeing, the requirements involved in design and manufacture of the vehicle would be the relevant ones. Under which every vehicle in that picture- and indeed anything in the US with less than 4 wheels- is a motorcycle.

Which type of license the driver needs would not be relevant at all-- Why would you care what state license the thing has?


Well, you were the one who claimed there were no 3 wheeled cars and then brought up an obscure bureaucratic definition that only applies to 2 countries in the world to explain why it would matter to the FSD computer.

I mean- FSDb is only actually deployed in those two countries. So seems pretty relevant?

But your desperation in trying to not admit you posted pics of motorcycles, not cars, continues to be amusing.
 
From the copyright notices, that data is sourced mainly from Google. There’s rumors that Tesla updates that. Perhaps by crowdsourcing.

Only the base maps are from google-- none of the nav/routing/detail is.


Don’t know how to get a quick fix, but hitting the right scroll wheel each time this occurs and intoning the words, “Bug report wrong speed limit should be 25 not 35” might help.

It won't. Those reports don't go anywhere- they stay local to the car and a service center can access them if you open a service ticket- all the report is doing is bookmarking the current car logs so they can see what it was doing when a problem happened- it won't help mapping issues at all.
 
The other problem is what speed should Tesla use when it doesn’t have any information or if there‘s a discrepancy between the database and the camera. It’s easy to say “obviously it should use the camera data,” but what about times where the sign is obscured by another car? In reality, humans use a combination of personal recall, actual signage and other visual cues. How do you program that?

It should do what most people do; look at what the cars around you are doing and flow with the traffic until the uncertainty is resolved.
 
  • Like
Reactions: GSP
Near miss of a bus. Now some people will claim that its clearly trying to go ahead of the bus based on the planner and won't hit it. Which is completely absurd. This adds to my point that making conclusions from planner tentacle watching is not a accurate thing to do.

9 mins 10 seconds
 
Now some people will claim that its clearly trying to go ahead of the bus based on the planner and won't hit it.

Also some people will claim that it wouldn't have applied the brakes.

I've had a few experiences like the above. FSDb angles itself to merge back into traffic, starts moving, and then slams on the brakes once it realizes the lane isn't clear. It's not comfortable, but I also wouldn't assume it would have hit the bus if not for the disengagement.
 
Maybe FSD-beta is just trying to mimic human drivers, that is to never even see motorcycles at all, or at the least to pay them no never mind, pull out in front of them, cut them off, etc. Unless of course, there are a lot of motorcycles, riders all wearing club jackets with logos like "Hecks Angles"........
 
  • Like
Reactions: FSDtester#1
It should do what most people do; look at what the cars around you are doing and flow with the traffic until the uncertainty is resolved.
This will be a difficult problem to solve. Many posted speed limits are not the speed actual drivers drive especially in neighborhoods. My neighborhood is posted at 30mph on all streets which is far too fast and nobody goes that fast.

Today the car sped up in an intersection which happened because the database reflected a speed limit increase in the intersection which is nuts. I wish Tesla would either tell all FSD testers how we can help by updating the relevant database (i.e.Tomtom) or tell everyone not to bother since it's a waste of our time.
 
Maybe FSD-beta is just trying to mimic human drivers, that is to never even see motorcycles at all, or at the least to pay them no never mind, pull out in front of them, cut them off, etc. Unless of course, there are a lot of motorcycles, riders all wearing club jackets with logos like "Hecks Angles"........
California is the only state where motorcycles can split lanes, which is really dangerous. Not sure why they don't just kill that law once and for all. Definitely have had a few near misses where traffic is going 20-30MPH on the freeway, and an opening comes up to make a lane change, but a motorcycle is splitting lanes at 50-60MPH. Check your blind spot, nothing there, start to make your lane change... where the hell did he come from?!?!?!?!?

I had a case once where the left lane was ending due to construction, so traffic was slowed down considerably as the merge was happening. It was my turn to merge over, and someone in the #2 lane made room for me, which was nice. I had my blinker on and started to merge over, and a motorcycle came by, splitting the #1 and #2 lane. I nearly hit him, and he got soooo pissed at me that he kicked my car twice, screaming at me, then took off. I was doing everything correctly, I had no choice but to merge, as the lane was ending with orange cones. My blinker was on, and everyone was merging smoothly. I was pissed off the rest of the day.
 
It should do what most people do; look at what the cars around you are doing and flow with the traffic until the uncertainty is resolved.


Tesla already got in trouble once for allowing FSD to break the law- most people speed, it's unlikely they'd ever adopt a "go whatever speed humans are" policy over sticking to map data and visible speed limit signs.
 
recall, actual signage and other visual cues. How do you program that?
I think program in recall, actual signage, and use other visual cues (like surrounding traffic speeds if it aligns with other sanity tests). It’s pretty easy to guess at approximate speed limits, given road width, conditions, etc. Should be easy, just send it to a neural net when it’s not clear, and certainly don’t wait for the first sign - you don’t need speed limit signs to know the speeds; there are standard speeds.

This proves my point that making conclusions from planner tentacle watching is not accurate.

It’s interesting. The visualizations lag by a half second or so, yet to the observer you do usually see a (seemingly) instant response to the tentacle part of them. Now of course, as long as the response does not come before the tentacle (definitely seems possible but I have not observed it on a small sample of video reviews), we can’t know whether the car is acting on something it decided to do up to 0.5 seconds before the tentacle moved.

In this video the wheel seems completely in sync with the tentacle.

It definitely seems odd to respond to something that lags by 0.5 seconds - but on the other hand the tentacle plots routes through cars - so maybe it has nearly zero latency while the visualizations have all the latency? It Is coming from a different source so seems conceivable.

Not sure what to make of it except that visualizations are completely useless (I never use them unless I am stationary and just want amusement).

4848D44E-41DF-4F47-9271-BB02A7C4CE52.png


 
Last edited:
  • Like
Reactions: Bladerskb
Tesla already got in trouble once for allowing FSD to break the law- most people speed, it's unlikely they'd ever adopt a "go whatever speed humans are" policy over sticking to map data and visible speed limit signs.

Tesla already has that policy. I can tell the car to automatically drive above the speed limit at all times. It’s a setting in the menus and it is applied even when driving with FSD.