AutoSteer nerfed in EU

[emmz0r]

green on Twitter

This is terribad!

Due to new local regulations, the limit of how far the steering wheel can turn while Autosteer is active has been adjusted. This may reduce Autosteer’s ability to complete sharp turns.

Additionally, to initiate Auto Lane Change, the turn signal must be engaged to the first detent (held partially up or down) and the lane change must start within 5 seconds of engaging the turn signal

https://www.unece.org/fileadmin/DAM/trans/main/wp29/wp29regs/2018/R079r4e.pdf

The system may exceed the specified value aysmax by not more than 0.3 m/s², while not exceeding the maximum value
specified in the table in paragraph 5.6.2.1.3. of this Regulation
...

Or is the wheel turning angle something else?

 

[J1mbo]

Seems to me that the regulatators want cars to slow down before making sharp turns, and this comment is there because Tesla cannot guarantee their system will do that.

 

[emmz0r]

Seems to me that the regulatators want cars to slow down before making sharp turns, and this comment is there because Tesla cannot guarantee their system will do that.

Are you absolutely sure about that? Why can't they garantuee that they will do something that in essence gives them more time to react and adjust?

"This may reduce Autosteer’s ability to complete sharp turns."

Sounds like it will beep and scream and disengange.

 

[J1mbo]

Are you absolutely sure about that? Why can't they garantuee that they will do something that in essence gives them more time to react and adjust?

"This may reduce Autosteer’s ability to complete sharp turns."

Sounds like it will beep and scream and disengange.

Exactly - it will do that when it hasn't slowed down enough to make the turn.

I am guessing this is most likely to happen on off-ramps with an inconsistent curve (and limited visibility of the curve), like this:

 

[verygreen]

and limited visibility of the curve

this (limited visibility) does not really matter in the real world. "Fog of War" is only a thing in videogames, in real world you can see entire map 100% of the time.

 

[paaltio]

Looks like Enhanced Summon will be completely forbidden because of this. So everyone in EU, write to your representatives. And with the EU election coming up, ask the candidates if they support autonomous vehicle innovation. Voting matters!

 

[RedBall]

As I see it, the tighter turning will be allowed as long as the speed of this turndoes not result in latteral movement of more than 3ms.

What constitutes a "tight turn" is totally dependent on the speed of travel.

Ive had a go below, but somebody with a better calculator may be able to pull my figures apart..

minimum desired radius of a 60km/h road as per this document is around 255m (less with more camber)
http://www.standardsforhighways.co.uk/ha/standards/dmrb/vol6/section1/td993.pdf

Circumference of that radius is 1602m.
At 60km per hour thecar would be covering 16m/s.
To cover 1/4 of the circumference (so latteral distance of one radius) would take 400.5/16 = 25 seconds.
therefor you will have covered 255m of latteral movement in 25 seconds, or around 10m/s, way over the 3ms.

to work it backwards, and see the maximum speed this bend could be taken at:

255m radius / 3m/s = 85
so you would need to take 85 seconds to travel this bend (1/4 the circumference)
so if 85 seconds = 1/4. 340s = entire 1602m circumference.
1602/340 = 4.7m/s. or 10.5 mph

If my figures above are correct, this most certainly will nerf autopilot. A lot.

 

[Eno Deb]

I don't get it. The document linked in the first posting is a UN regulation. What does that have to do with the EU?

 

[RedBall]

I don't get it. The document linked in the first posting is a UN regulation. What does that have to do with the EU?

It appears, from what I can gather, that there is no defined standard for vehicle type approval when it comes to autonomous stuff. But within the EU they have now used this UN document as a base for their type approval.

 

[verygreen]

the variable that controls this behavior is named GUI_applyEceR79 so I guess this is the correct regulation to look at.

 

[Eno Deb]

It appears, from what I can gather, that there is no defined standard for vehicle type approval when it comes to autonomous stuff. But within the EU they have now used this UN document as a base for their type approval.

It appears what this regulation does is allow reciprocal approval in all countries that adopt it, i.e. instead of getting approval for every country individually a carmaker only has to go through the process once in one signatory country. If it only affects European countries it seems the US has not adopted it?

 

[UltimateX]

Long time lurker here.

As a matter of Engineering i have to step in with some physics. The limitation of 3m/s2 lateral acceleration allows easily to calculate the limit speed for a given turn radius.
a= v sqared / r
v = square root ( a r )

This results in:
Max turn radius r = 12 m -> 6 m/s = 21.6 km/h
Normal tight turn r = 20 m -> 7.8 m/s = 27.9 km/h

If autopilot takes these turns with that speed I would call it spirited driving. I will be quite glad when my Model X will cause me to grab onto the door handles in city travel.

 

[J1mbo]

this (limited visibility) does not really matter in the real world. "Fog of War" is only a thing in videogames, in real world you can see entire map 100% of the time.

Are they using the map for path planning? Thought they'd given up on that...

 

[TexasEV]

In other words, were not going to allow Tesla to do this until the German auto companies catch up.

 

[verygreen]

Are they using the map for path planning? Thought they'd given up on that...

they absolutely use map to determine if they need to reduce speed for upcoming road curvature.

 

[emmz0r]

they absolutely use map to determine if they need to reduce speed for upcoming road curvature.

So they haven't moved to a heuristics based model on the path prediction yet?

 

[verygreen]

So they haven't moved to a heuristics based model on the path prediction yet?

They do that when there is no map data. But it's kind of late and people perceive it as sudden braking. They do it much more neatly with the map data in place.

 

[paaltio]

They do that when there is no map data. But it's kind of late and people perceive it as sudden braking. They do it much more neatly with the map data in place.

Is there something specific in the map data in some areas that allows it to do that? AP certainly doesn't seem to brake in time here in Finland. Instead it drives way too fast to a tight turn and then brakes at the last minute when the turn is already underway.

 

[I.a.n]

If autopilot takes these turns with that speed I would call it spirited driving.

I thought that when I first saw the figures - they are way above the 1.45m/s^2 to 2.45m/s^2 that seated rail passengers find "comfortable" - Passenger Stability Within Moving Railway Vehicles: Limits on Maximum Longitudinal Acceleration

the variable that controls this behavior is named GUI_applyEceR79 so I guess this is the correct regulation to look at.

However there is a second element to that regulation:

(a) The steering control effort necessary to override the directional control provided by the system shall not exceed 50 N;

I wonder if the car has sufficient sensors to separate the torque required to turn the wheel against the road force and the user "fighting" for control? I.e. the steering motor probably has a electrical current measuring device on it, and the more current it uses the the larger the force is being exerted.

If the car is no longer allowed to exert more than 50N (is that the same as hanging a 5Kg/11lb weight from the wheel?) before disengaging AP, but it cannot distinguish between the force the driver is exerting and the forces from the wheels on the road (frictional & reactive to the turning I presume), then that may impose a lower limit on the tightness of the turn?

Does anyone know what sensors are available on the steering column?

Absent of a strain gauge on the steering column, I presume the software could estimate the expected current required to achieve a given turn radius (with a margin of error for wet or dry roads/concrete vs asphalt/winter vs summer tyres) and disengage if the force is more than 50N away from the estimate. It might even take two software updates - you can imagine one which records the maximum and minimum current for any given turn (and uploads that to the mother ship), and the next that uses that with some margin of error to disengage. Potholes might be a problem though (presumably they cause a spike in current, just like a driver tugging on the wheel would).

 

[fmonera]

Is this rule really problematic for the future of EAP/FSD? It seems to me just that UN rule requires the car to take a curve at an adequate speed, so TESLA just needs to improve it's heuristic system to slow down before taking a curve.

Until all this is clarified, I am halting my ModelX updates. 2019.12.1.2 is good enough until we have NoA in Spain which is the only thing that could make me accept the UN-rule regression.