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

AP camera distance sensing assumes vehicle sizes, sometimes incorrectly

This site may earn commission on affiliate links.
I was driving next to a small trailer pulled by pickup truck and noticed that the AP cluster display was displaying the car as a big-rig and it was showing farther ahead my car than it really was (even when it was right next to the car in front of me, it appeared ahead it). Based on my observations, it appears that the distance estimation for cars in adjacent lanes is done using the AP camera AND is based on some some assumed sizes of vehicles (drawback of mono-vision). That trailer, which looked almost identical to the pictured below (same brand), was mistakenly recognized by AP as a big rig trailer, but because it is smaller, AP thinks it's farther out. I'm not sure how that would affect lane intrusion detection, but it could cause issues.

TW101.jpg
 
I was driving next to a small trailer pulled by pickup truck and noticed that the AP cluster display was displaying the car as a big-rig and it was showing farther ahead my car than it really was (even when it was right next to the car in front of me, it appeared ahead it). Based on my observations, it appears that the distance estimation for cars in adjacent lanes is done using the AP camera AND is based on some some assumed sizes of vehicles (drawback of mono-vision). That trailer, which looked almost identical to the pictured below (same brand), was mistakenly recognized by AP as a big rig trailer, but because it is smaller, AP thinks it's farther out. I'm not sure how that would affect lane intrusion detection, but it could cause issues.

View attachment 114530

Did you report this to Tesla?
 
Did you report this to Tesla?

Not yet. I need to write it up with some potential solutions before sending it in. So far my solution suggestions are:
1. Identify the license plate location, then based on its size determine whether the assumed size of the vehicle is correct. This would be unique per region since license plates vary between regions.
2. If #1 fails, or in addition to #1, determine the width of the lane that the MS is traveling in (knowing the width of MS and AP camera angles), then compare the size of the other vehicle to the size of the lane it is traveling in, assuming their lane is close in width to your lane.

Not sure what else can be done without stereo vision. If Tesla uses deep learning, then they could add a bunch of such trailers and other vehicles that look like larger vehicles into their deep learning training.
 
I did did not think the camera was involved at all. I thought it was just radar based. Mine always detects a fiberglass school bus body as a car, rather than a big rig. OTOH, it always detects my Honda Fit that I park directly behind, within 4 inches, as a truck in my garage. Go figure.

I was driving next to a small trailer pulled by pickup truck and noticed that the AP cluster display was displaying the car as a big-rig and it was showing farther ahead my car than it really was (even when it was right next to the car in front of me, it appeared ahead it). Based on my observations, it appears that the distance estimation for cars in adjacent lanes is done using the AP camera AND is based on some some assumed sizes of vehicles (drawback of mono-vision). That trailer, which looked almost identical to the pictured below (same brand), was mistakenly recognized by AP as a big rig trailer, but because it is smaller, AP thinks it's farther out. I'm not sure how that would affect lane intrusion detection, but it could cause issues.
 
I did did not think the camera was involved at all. I thought it was just radar based. Mine always detects a fiberglass school bus body as a car, rather than a big rig. OTOH, it always detects my Honda Fit that I park directly behind, within 4 inches, as a truck in my garage. Go figure.


If you cover the camera (or if it frost over as seen in Tesla Motors: AutoPilot Freeze ) you will not see any cars on the AP display.
 
Did you report this to Tesla?

Not yet. I need to write it up with some potential solutions before sending it in. So far my solution suggestions are:
1. Identify the license plate location, then based on its size determine whether the assumed size of the vehicle is correct. This would be unique per region since license plates vary between regions.
2. If #1 fails, or in addition to #1, determine the width of the lane that the MS is traveling in (knowing the width of MS and AP camera angles), then compare the size of the other vehicle to the size of the lane it is traveling in, assuming their lane is close in width to your lane.

Not sure what else can be done without stereo vision. If Tesla uses deep learning, then they could add a bunch of such trailers and other vehicles that look like larger vehicles into their deep learning training.

It's great that you want to present the problem along with potential ways for Tesla to solve the problem. But frankly I think letting Tesla know about the issue is far more important. The likelihood of Tesla needing any of your proposed solutions is pretty low. What I mean is either they would come up with whatever you come up with on their own or they'd come up with something better. The most important thing is to alert them to the fact that an issue exists.

Thanks!