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

Collision Avoidance warning at dip in road

This site may earn commission on affiliate links.

RDoc

2021 Prerefresh Model S
Aug 24, 2012
2,805
1,767
Boston North Shore
I was on a local road slowing for a stop sign, brakes and regen, when the collision warning suddenly came on. The road was going down into a small dip with the stop sign at the opposite top with a car stopped at it. There was plenty of space and I was slowing down easily quickly enough to stop very safely, actually I eased to a stop.

My guess is that the front radar saw the car, but then got confused by the dip in the road and started reading the rise on the other side from where I was as the distance to the stopped car and panicked. Nothing untoward happened, but it is a situation the system didn't handle correctly.
 
I was on a local road slowing for a stop sign, brakes and regen, when the collision warning suddenly came on. The road was going down into a small dip with the stop sign at the opposite top with a car stopped at it. There was plenty of space and I was slowing down easily quickly enough to stop very safely, actually I eased to a stop.

My guess is that the front radar saw the car, but then got confused by the dip in the road and started reading the rise on the other side from where I was as the distance to the stopped car and panicked. Nothing untoward happened, but it is a situation the system didn't handle correctly.

Might have warned you because of the combination of dip and distance.
Downslope makes it harder to slow down. (And don't I hate it when reduced speed is on a steep downslope instead of putting it on a flatter bit or a crest.) "Oh damn I'm going downhill and there's a car right there! Aaaaarrrrggghhh!"
 
Might have warned you because of the combination of dip and distance.
Downslope makes it harder to slow down. (And don't I hate it when reduced speed is on a steep downslope instead of putting it on a flatter bit or a crest.) "Oh damn I'm going downhill and there's a car right there! Aaaaarrrrggghhh!"

My driveway dips before joining the road, which is just coming off a curve and is banked. I would guess that my alarm goes off about once in five times, usually when I am approaching the road quickly (my drive goes downhill to the road), when I hit the brakes to check for traffic, then "BEEP BEEP BEEP" just before the wheels nudge the nose upward again. If I go slower, no problem. If I go faster, it grouches. It thinks I am driving into a wall at speed.

I have also noticed that the dip is worse near the edge of the drive, and if I stay more centered (drive is 28 feet wide) there is no problem. It doesn't bother me, since I know what causes it and how to keep it calm as I hot rod onto the road.

It does give a little excitement to the test ride when I have passengers, though. "What's THAT??"
 
Because the dip can be interpreted as a dip, and not an object.

Maybe I'm not envisioning the dip correctly :confused:

A stereoscopic camera will help make a 2D scene 3D, I understand that. The way I envision the dip is something like this (but less extreme):
dip-road-17369478.jpg


But that's a higher from the ground view than the Model S camera currently is (or so I assume?), so I'm not sure how much depth perception a 2nd camera will give you for a dip.

The way I see it [for this particular case. there are tons of places where the following statement isn't true] is that if you can do this with 2 cameras, you should be able to do it algorithmically with 1 camera with a fairly high degree of certainty. Though it's been a while since I've done computer vision, so I'm sure I forgot something?
 
Maybe I'm not envisioning the dip correctly :confused:

A stereoscopic camera will help make a 2D scene 3D, I understand that. The way I envision the dip is something like this (but less extreme):
dip-road-17369478.jpg


But that's a higher from the ground view than the Model S camera currently is (or so I assume?), so I'm not sure how much depth perception a 2nd camera will give you for a dip.

The way I see it [for this particular case. there are tons of places where the following statement isn't true] is that if you can do this with 2 cameras, you should be able to do it algorithmically with 1 camera with a fairly high degree of certainty. Though it's been a while since I've done computer vision, so I'm sure I forgot something?

Humans can barely do this right. Remember the blue/black white/gold dress thing? It's not clear from the image what's going on, you apply some intuition to it based on previous experience to guess what it is, and it may be right. The only thing that's clear from the image is that *something* is going on, which in real life might even just be a shadow. 3D camera will be able to triangulate distance, and therefore you have enough information to see if it's a dip, a bump, or a shadow.
 
I certainly don't know how TACC and AP are programmed, but they don't seem to very sophisticated in assessing what's around the car. I wonder if a useful software architecture would be to try to build up a 3D virtual model of what's around the car, constantly updating it based on what the sensors are perceiving and the car's positional history.

In this case, the car would be able to recognize that at the top of the dip, it saw the stopped car at a distance, but it couldn't see the road in front of the Tesla. Then, as it entered the dip, the stopped car didn't move, based on it's position and the change in position of the Tesla, but now it started seeing the road appear as reported by a combination of the radar and mono camera, If it was building a model of all this rather than just responding to something that suddenly appeared, it might have a better chance of interpreting what was in front more usefully.
 
A long time ago when I had an Acura RL and followed those forums, there were long threads of the radar warning of imminent doom where none was actually present. The most interesting to me was a down hill grade with slightly raised train tracks crossing the road at the bottom. Due to the angle of the car facing down, and the slightly raised steel tracks, the radar would obviously reflect off the steel and cause the car conniptions that it was going to crash into something solid.