New radar mode discussion

[thegruf]

As I understand it, the roll out of 8.0 enhances the radar use to a scanning mode to build up a 3D model of the world it sees in a very similar way to LIDAR operation but at radar frequencies, this now includes the ability to "see" stationary objects. Damn clever stuff for sure.

Initially with v8.0 the radar is now running in this mode to acquire data that is sent back to the mothership to generate a whitelist of radar reflections from road signs, bridges and other stationary objects.

At this stage though, the changes to AP have not been implemented to work with this new use of the radar, precisely because of unwanted false detection events with the resultant bucketload of complaints about false braking.

Tesla have not stated when the new mode will become active, (I guess 8.1 at the end of the year?) however there are some points to consider when it does go active

- presumably it cannot go active on all roads at once - some may not have sufficient previous data
- how does the driver then know if the AP is in standard mode or "advanced" mode and will this be indicated to the driver?
-What if there is an obstruction under say a bridge ( I am reminded of the rear end shunt when a driver collided with a van at the edge of the road close to a bridge). If the whitelist is gps based then it might not be able to detect this, if it is data based and for example could subtract just the bridge "image" then surely this is a ton of data.
-if the whitelist data is large, then Tesla will have to ration what the car receives to the local area, what if you go outside of this area on a trip? What if you have no mobile signal to obtain updated datasets?

No doubt Tesla have this all thought through, I am curious nonetheless

 

[calisnow]

Lots of good questions OP and few answers. There is another issue - stationary cars in front of you which were not moving when they came into view. AP 7.1 and 7.0 famously had trouble with these - what about 8.0?

A stationary car on a highway will never be whitelisted because it is only a temporary object - so the question is - does 8.0 deal with stationary cars any better than 7.1 and 7.0?

I will tell you this much - 8.0 seems to me to handle the traffic around me better than 7.1 did in several subtle but noticeable ways. It also routinely sees the car in front of the car in front of me and shows me on the screen.

The question about stationary cars though, still is a mystery.

 

[Vad42]

So on two occasions now, I was autopiloting at 50-60 mph with no cars in front of me. Empty road and nothing on radar, just lines. And come up to dead-stop traffic. I mean not moving at all. I was carefully monitoring the radar display, but cars where not appearing and my car wasn't braking. In both cases I chickened out and pressed brake myself.

It was already later than I would normally start slowing down in such case. I didn't want to get into emergency breaking situation, by me or autopilot. It's certainly possible it would have been fine, but I was afraid to truly test it. Even if it would have been fine, it would be jarring and uncomfortable. Anyone else observe it behavior in such situation? How many feet from the car ahead of you should it start braking for this to not become emergency event?

This is with 8.0, as I have almost no autopilot miles on 7.1 to compare.

 

[thegruf]

My concern re possible switching of AP modes on different roads is variance of behavior.

We can all get used to how a system works so long as it works consistently, but if one minute AP is super effective seeing everything and reacting accordingly and we dial our responses into that, then I turn onto a different highway and it reverts to its current (not exactly shabby) mode due to lack of whitelist data for that road, I can see this causing some consternation.

Hopefully somebody has some more ideas how this can work

 

[MarkS22]

You bring up one of my biggest questions. How detailed are the radar signatures? If a bridge or sign is whitelisted, what if a large object is stopped at those coordinates?

For example, in the fatal Florida accident, what if that intersection had an overhead highway sign over that exact location? Is the whitelisting/radar able to say "this radar return shape is whitelisted, but this is a new one... slam on the brakes."

My concern on this not being possible is that, if it could accurately tell the difference between a bridge/sign/car/truck, why even need whitelisting?

 

[Electroman]

You bring up one of my biggest questions. How detailed are the radar signatures? If a bridge or sign is whitelisted, what if a large object is stopped at those coordinates?

For example, in the fatal Florida accident, what if that intersection had an overhead highway sign over that exact location? Is the whitelisting/radar able to say "this radar return shape is whitelisted, but this is a new one... slam on the brakes."

My concern on this not being possible is that, if it could accurately tell the difference between a bridge/sign/car/truck, why even need whitelisting?

I have the same concerns too. Also I feel it is impractical to maintain a database of ever changing stationery objects in your white-list.

I would like to understand why identifying a stationery object in your path - between the two lines - a frontal area with a width of your lane and the height of your vehicle is that difficult? I guess it gets complex in curves, where the car doesn't realize the lanes are curving until it is only about say 20 feet before the lanes starts curving, and by that time the car would have already made a false positive identification of a stationery object, which happens to be in front of you but outside your lanes.

 

[brkaus]

I have the same concerns too. Also I feel it is impractical to maintain a database of ever changing stationery objects in your white-list.

I would like to understand why identifying a stationery object in your path - between the two lines - a frontal area with a width of your lane and the height of your vehicle is that difficult? I guess it gets complex in curves, where the car doesn't realize the lanes are curving until it is only about say 20 feet before the lanes starts curving, and by that time the car would have already made a false positive identification of a stationery object, which happens to be in front of you but outside your lanes.

And hills/changes in elevation.