Merging traffic and autopilot

[spiderbro]

I think it's been decently documented since AP1 and the issue still continues with AP2 hardware - how does Tesla plan to deal with merging (and diverging) traffic?

AP does not see merging traffic quickly enough and (probably less importantly) does not do well with diverging traffic either. For clarification I am referring to:

1) Merging traffic - Tesla will not slow down/brake until much too late before it sees an incoming car and when it does braking is very sudden.
2) Diverging traffic - When the front car leaves the lane Tesla will abruptly speed up and then abruptly brake when it picks up the next car.

Can/will improvements with cameras and additional sensors be adjusted to deal with this? It seems like a fundamental facet of driving that needs more attention.

 

[chillaban]

It looks like the current direction for AP2/EAP is the same Autosteer+TACC strategy we see in AP1. Of course, how FSD will be is a different story.

From a theoretical point of view the car has cameras pointed in every direction and has enough data to understand merging patterns. Whether or not it will be effectively used is an entirely different story. There's a lot that AP1 should theoretically be able to do given its sensors but does not do well / at all.

 

[spiderbro]

At the moment AP2 doesn't even render adjacent cars so I am hoping after adjacent cars are added there will be some progress as I don't see AP going far without merging traffic being addressed more appropriately. It is a big safety issue.

 

[Derek Kessler]

I have no doubt that this is high on the priorities list for getting absolutely right. Remember, we're still in the very early days for AP2 hardware and likely years away from a completely public roll-out of Full Self-Driving. That AP2 doesn't render adjacent cars right now shouldn't be a surprise — they're still working on getting it on par with AP1!

 

[Electroman]

The problem is not about having the ability to see merging traffic. It is the difficulty in identifying that cars in the next lane are actually merging. Especially in onramp scenarios, it is not easy to realize that these are traffic that are not going to continue in the next lane but are actually merging.

 

[Sawyer8888]

Merging traffic - Tesla will not slow down/brake until much too late before it sees an incoming car and when it does braking is very sudden.

Now, over a year later, the inability of AP2 to recognize merging traffic persists.

This happened to me this morning in relatively slow stop-and-go traffic (set to one car length). After coming to a halt on the highway, my car took it's sweet time in deciding it was ready to go when traffic started moving again <-- typical and annoying. Naturally, the cars merging from the on-ramp to my right thought I was giving them room to enter my lane. Not so. The vehicle to my right put their blinker on and slowely started merging in front of me. What did my car do? It began to quickly accelerate to catch up to the other car that was now several lengths ahead, completely ignoring the new car entering my lane. I came within a foot or two of hitting the merging vehicle which forced me to cancel AP and caused that car to quickly pull back to the right since they thought an accident was imminent - which it was. Then another motorist honked at me for my apparent selfishness.

What a pain. With all these sensors and cameras, it still cannot recognize another car making a simple turn. I just wrote to [email protected] describing the situation.

I have no doubt that this is high on the priorities list for getting absolutely right.

Good thing you didn't place any bets on it.

 

[Derek Kessler]

Good thing you didn't place any bets on it.

Never gave a time for when — but it's definitely still something Tesla would want to be sure they've gotten right before deploying.