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

V9 Auto Lane change and high speed traffic

This site may earn commission on affiliate links.

widodh

Model S 100D and Y LR
Moderator
Jan 23, 2011
6,861
2,840
Venlo, NL
As I live in the Netherlands I drive in Germany quite often. Yesterday I got back from a 2.000km roadtrip over the German Autobahn where speeds can easily exceed 250km/h.

My S100D is still on v8.1, but I was wondering how the Auto Lane change in V9 can work with high speed differences between cars.

On a 3-lane highway it's not uncommon that you see these speeds:

- Left: 160 ~ 200km/h
- Middle: 110 ~ 140km/h
- Right: <110km/h

So if I'm driving in the middle lane and I'm on 'Mad Max' mode, can the cameras on Model S/X detect a car coming in the left lane with 180km/h before performing the lane change?

As a driver I look in the mirror a couple of times before I'm sure I can change lanes.

This might be less of an issue in the US, but still, you can have somebody speeding and you don't want to drop your car in front of them.

Can the cameras on Model S see far enough behind to detect this?
 
If the software is working correctly (remember this will be all "beta" software), it should be using the sensors to identify objects around the vehicle, including moving vehicles ahead, behind and to the sides - and estimating the relative speed of each vehicles and the spacing between vehicles - and using that data to control lane changing, like TACC uses this information to control vehicle speed.

Until Tesla has been able to test this on many vehicles on many roads on many situations, this feature (like any of the AutoPilot features) should be treated as "beta", and the driver shouldn't assume the software will always take the correct action - and should be prepared to take control at any time.

Wouldn't recommend using "Mad Max" mode, until there's confidence the software is working correctly using the more conservative settings.

Since this is a "driver assistance" feature - if there is an accident because the software makes a mistake, the responsibility will be with the driver - and not Tesla...
 
If the software is working correctly (remember this will be all "beta" software), it should be using the sensors to identify objects around the vehicle, including moving vehicles ahead, behind and to the sides - and estimating the relative speed of each vehicles and the spacing between vehicles - and using that data to control lane changing, like TACC uses this information to control vehicle speed.

Until Tesla has been able to test this on many vehicles on many roads on many situations, this feature (like any of the AutoPilot features) should be treated as "beta", and the driver shouldn't assume the software will always take the correct action - and should be prepared to take control at any time.

Wouldn't recommend using "Mad Max" mode, until there's confidence the software is working correctly using the more conservative settings.

Since this is a "driver assistance" feature - if there is an accident because the software makes a mistake, the responsibility will be with the driver - and not Tesla...
I fully understand your answer and I agree.

But I was mainly wondering how Tesla would be able to detect this technically as to me it seems that sensors at the back can't see far enough to see a vehicle approach with a high speed difference.

If I'm driving on the Autobahn I will probably not use this feature.
 
Other manufacturers claim LIDAR is needed to properly detect all objects that could impact driving.

Since AP2 was introduced 2 years ago, Tesla has claimed this can be done reliably with the combination of a forward facing radar, 8 cameras and the proximity sensors.

To operate safely, the AP software must be able to detect all objects, in 360 degrees around the car, and how they may impact the car as the car is moving on its intended path. In real time, the AP software is essentially building a 3D model of the surroundings, classifying everything surrounding the AP vehicle.

For each object, the software must be able to determine the object is present, classify the type of object (stationary/moving, car/truck/motorcycle, pedestrian, traffic sign, potential obstruction, …), location (where it currently is relative to the AP vehicle, how close/far it is, the relative 360 degree position), speed/direction (how it's moving relative to the AP vehicle), size (how much space it is taking), …

Tesla initially relied on Mobileye for object detection - and replacing it caused a significant delay in getting AP2 to work well using only the front sensors.

Being able to use the side cameras to reliably detect objects on either side of the AP vehicle is a critical next step on the path to achieving FSD...
 
  • Informative
Reactions: fmonera
In countries with very high rate of
Speed cars in the left lane (autobahn) I would think autolane change to the far left should not be used. Use your mirrors and judgement for that and use ULC for other safer scenarios (speed controlled roads, etc).
 
In countries with very high rate of
Speed cars in the left lane (autobahn) I would think autolane change to the far left should not be used. Use your mirrors and judgement for that and use ULC for other safer scenarios (speed controlled roads, etc).
The current cars are ostensibly ready for full autonomy with the hardware they have.
They will need to handle dealing with speeding cars.
 
As I live in the Netherlands I drive in Germany quite often. Yesterday I got back from a 2.000km roadtrip over the German Autobahn where speeds can easily exceed 250km/h.

My S100D is still on v8.1, but I was wondering how the Auto Lane change in V9 can work with high speed differences between cars.

On a 3-lane highway it's not uncommon that you see these speeds:

- Left: 160 ~ 200km/h
- Middle: 110 ~ 140km/h
- Right: <110km/h

So if I'm driving in the middle lane and I'm on 'Mad Max' mode, can the cameras on Model S/X detect a car coming in the left lane with 180km/h before performing the lane change?

As a driver I look in the mirror a couple of times before I'm sure I can change lanes.

This might be less of an issue in the US, but still, you can have somebody speeding and you don't want to drop your car in front of them.

Can the cameras on Model S see far enough behind to detect this?

I'm highly skeptical that the current sensor suite will ever work safely in this situation. I'm one foot in the lidar camp... there are a lot of oddball situations, where I think that the camera only solution is going to struggle to be reliable enough to be safe.
 
As I live in the Netherlands I drive in Germany quite often. Yesterday I got back from a 2.000km roadtrip over the German Autobahn where speeds can easily exceed 250km/h.

My S100D is still on v8.1, but I was wondering how the Auto Lane change in V9 can work with high speed differences between cars.

On a 3-lane highway it's not uncommon that you see these speeds:

- Left: 160 ~ 200km/h
- Middle: 110 ~ 140km/h
- Right: <110km/h

So if I'm driving in the middle lane and I'm on 'Mad Max' mode, can the cameras on Model S/X detect a car coming in the left lane with 180km/h before performing the lane change?

As a driver I look in the mirror a couple of times before I'm sure I can change lanes.

This might be less of an issue in the US, but still, you can have somebody speeding and you don't want to drop your car in front of them.

Can the cameras on Model S see far enough behind to detect this?
It's a pretty common sight to see >110MPH ghetto cars like BMW or Honda passing me. I almost never see luxury or middle class vehicles going that fast.
 
Human drivers use two video sensors that can rotate roughly 180 degrees, using three mirrors to gain a limited view of objects to the side and rear of the vehicle.

The AP2 cameras should be providing at least this much information, probably much more, in addition to the front-facing radar and proximity sensors.

Having 360 degree radar or lidar would clearly provide even more data - and should make it easier for the software to detect objects and their relative size, location and speed.

Though since humans don't have those sensors and rely instead on more limited visual data, it seems possible Tesla should (eventually) get AP working well for EAP/FSD, assuming the AP sensors are sufficient to provide a 360 degree view at least as good as a human driver has.

Tesla appears to have the front-facing cameras/radar working now - operating at least as good as Mobileye's AP1 system. Will be interesting to see what Tesla provides in V9 - and how much progress they are making with the other sensors.
 
The cameras have a reasonable field of view of neighboring lanes, but recognizing cars that are substantially far away with large delta speeds will be very tricky.

Unrestricted autobahns are a tiny percentage of global Tesla usage, and therefore is very much an edge case. I think the short answer is don't expect it to be safe to use on an unrestricted autobahn for a long time.

Keep in mind also that the V9 release will just suggest lane changes, it won't make them without user intervention, so the driver will be expected to confirm that the lane is clear before a lane change is initiated.
 
The cameras have a reasonable field of view of neighboring lanes, but recognizing cars that are substantially far away with large delta speeds will be very tricky.

Unrestricted autobahns are a tiny percentage of global Tesla usage, and therefore is very much an edge case. I think the short answer is don't expect it to be safe to use on an unrestricted autobahn for a long time.

Keep in mind also that the V9 release will just suggest lane changes, it won't make them without user intervention, so the driver will be expected to confirm that the lane is clear before a lane change is initiated.
I agree that it's a short percentage, but if they enable 'auto lane change' in Europe it's going to be interesting in Germany!

I don't think the current hardware can see far enough in the other lane to make a good judgement.
 
Humans are detecting objects on either side of the vehicle using considerably less data than AP has available from the side/rear cameras and proximity sensors.

Reliably detecting the vehicles on either side will be challenging using primarily the cameras - though as long as the cameras are able to view objects with enough resolution, wide enough angle, and at near to far distances - AP should be able to do as well as a human, and should be able to do much better, since the cameras are constantly viewing the side areas, whereas humans are only glancing periodically to the sides.

It took Tesla much longer than they (Musk) expected to get AP2 object recognition working well enough to get AP2's EAP operating at close to AP1 levels (with Mobileye), so we shouldn't be surprised if it takes Tesla a while to get the side cameras fully operational.