How much will original Autopilot improve

[suraj1194]

Interesting Suraj. What does yours do when someone cuts in in front of you? Even though it would often be sufficient to just ease off the accelerator a bit, mine slams on the brake and slows down to where the car behind has to brake hard was well. Then it takes for ever to get back up to speed , opening an even wider gap in front, thus inviting more cars to cut in, etc. In those situations I tend to just turn it off. It didn't used to do that before the last iteration.

This is true. The algorithm is biased to immediate action when it detects forward obstruction, either through confusion about cars in adjacent lanes as mentioned above, or someone cutting in.

I typically keep follow distance at 4 for freeways speeds but tend to cut that in half when traffic gets worse. Maybe it's my own imagination, but the intensity of the braking is less severe when it's not 'trying to defend a large separation distance ahead'. I don't mind that it builds up speed again rather slowly. Earlier, as I mentioned too, it used to accelerate again rather quickly, which was rather difficult on passengers.

In summary it appears that the algorithm is written to bias to immediate action on any perceived forward obstruction but retains the accelerate again slowly bias for when it's again trying to make up for the too much space it let go when it braked. This could be improved by tracking relative velocity of the car cutting in and denoting it as less of a collision threat, and therefore not brake quite so hard and give up too much space, and instead brake judiciously and re-establish the prior separation distance.

From a programming perspective, this may be harder than described, because the AI needs to distinguish between immediate collision threat and a mere lane intrusion it can make up for by slowing a little. It also needs to account for how far ahead of the car the intrusion is and control the braking. I don't know if the decision tree for this is trivial, especially when the cost of making the wrong choice (e.g. the intruding car and everything in front comes to a sudden stop) can be severe...

 

[Snerruc]

This is true. The algorithm is biased to immediate action when it detects forward obstruction, either through confusion about cars in adjacent lanes as mentioned above, or someone cutting in.

I typically keep follow distance at 4 for freeways speeds but tend to cut that in half when traffic gets worse. Maybe it's my own imagination, but the intensity of the braking is less severe when it's not 'trying to defend a large separation distance ahead'. I don't mind that it builds up speed again rather slowly. Earlier, as I mentioned too, it used to accelerate again rather quickly, which was rather difficult on passengers.

In summary it appears that the algorithm is written to bias to immediate action on any perceived forward obstruction but retains the accelerate again slowly bias for when it's again trying to make up for the too much space it let go when it braked. This could be improved by tracking relative velocity of the car cutting in and denoting it as less of a collision threat, and therefore not brake quite so hard and give up too much space, and instead brake judiciously and re-establish the prior separation distance.

From a programming perspective, this may be harder than described, because the AI needs to distinguish between immediate collision threat and a mere lane intrusion it can make up for by slowing a little. It also needs to account for how far ahead of the car the intrusion is and control the braking. I don't know if the decision tree for this is trivial, especially when the cost of making the wrong choice (e.g. the intruding car and everything in front comes to a sudden stop) can be severe...

Strange, mine seems to recognize a car coming into my lane and just give it room. I'm set on 7 ( and that is considered too close by some people here who will slow down and pull off.). I wish there was an eleven.

 

[suraj1194]

Strange, mine seems to recognize a car coming into my lane and just give it room. I'm set on 7 ( and that is considered too close by some people here who will slow down and pull off.). I wish there was an eleven.

Well, this is probably a function of the nature of traffic. I don't suppose the traffic in Kona is anything like the Bay Area. Setting 7 really wouldn't work in rush hour here; it would be an open invitation for everyone and his dog to butt in between me and the car in front

 

[3mp_kwh]

* When the car in front kept me from going at preset speed and then it changed lanes opening up my lane, the car used to bolt forward like a jackrabbit (sort of). Not anymore. The acceleration is smooth, starting out slow and building momentum, rather than jerk forward.

Simultaneous with this change, I notice the TACC follow distance has substantially narrowed to between 2.5-4 car lengths, over the full 7 selectable settings. I wish that could be restored. The current "1" car length is too far not to invite being cut off by drivers in adjacent lanes, during stop & go. Conversely, the "7" setting was useful when you see traffic ahead and want AP to get going on the braking earlier, less aggressively or without having to disengage AP.

Otherwise, the only other recent AP1 frustration has come from the car signaling it can see painted lines (blue), yet doing so immediately as it drives across them and off the road. That happened to me on a rare side-street 25mph use of AP1, and cost me rim damage and two PS2s. If I had seen the car showing me it could not see the painted lines, I wouldn't at that moment have been changing to my sunglasses. That was all it took (18" shoulder, then rocks...). It seems crazy that if the animation display can show the car driving over the blue lines, that the auto-steer wouldn't be attuned to avoid doing just that. My hands weren't free, but my eyes watched it happen.

 

[Lucky_Man]

As an owner of a HW 1 car, I'm of course curious about how much it will improve with future software.
If it is self learning at no cost to Tesla, one would assume that it would improve. Has anyone seen evidence of self learning though? I have not. For example there is a road in my neighborhood where the car always selects the turn lane while driving in the straight ahead lane, blinkers off. When the car initiates the turn, I usually take over the wheel and steer the car back to the straight ahead lane, then turn on AP again. 8 months of doing this taught the car(s?) nothing.
So I wonder if, A; self learning refers to only non location based driving algoritms or B: self learning refers only to the AP2 neural network? Is there any learning and knowledge carry over from AP2 learning to AP1 perhaps?

If AP1 is then built on manual programming, the reasons for diverting coding resources from current and future products would be customer satisfaction and resale value. So is Tesla doing it? Or is it in fact self learning? Let the speculation begin...

Hey!, I noticed the same problem with our '15 AP1. Was only slightly concerned at the time. However, we just returned from the Black Hills rally in Custer, SD. This route was mostly through Wyoming. I was surprised when this 'problem' went away in Wyoming!

After discussing this at the Lusk SC, another driver from CO said it was because their State DOT paints 'dashed line' lane markers on the exit ramps. When we got back to CO, I noticed these dashed lines were NOT present in CO.

I plan to make a comment on this issue to our CDOT rep at the next meeting. With autonomous driving on the horizon, I know CDOT wants to keep up with evolving standards.

 

[LastGas]

Confused. Does AP1 navigate to freeway interchange ramps automatically based on the nav route????

Musk said it would with 8.1, but I have never seen this behavior. The Owners Manual, since 7.1, says that the car will slow down in advance of an exit ramp when the turn signal is on, but I have never seen that either.

 

[LastGas]

Musk tweeted in December:

Michael Wang:

"will we see continued improvements to AP HW1 after the 8.1 update?"

Musk:

"Yes"

and again in January:

"There will be many improvements to HW1 over the course of the year"

 

[Olle]

The problem with taking the particular on the straight that I initially complained about seems to have gone away. I drove on that road the other day and the car stayed in the straight lane for the first time. Yay!

 

[USMC_Mustang]

I'd be happy to receive improved integration between the front (optical) camera and the front radar. If we recall that Tesla transferred AP steering control from the camera to the radar in order to enable distance detection. I'm my opinion, this change diminished the quality and accuracy of AP highway driving. The car now too frequently vectors out of its lane (left or right) with no apparent reason and decelerates strongly when going downhill (e.g., the angle between the road surface and the radar closes, confusing the AP computer). I'm sure there must be room for improvement in utilizing the best features of the camera and radar together to enhance AP1's capabilities.

 

[David29]

Interesting Suraj. What does yours do when someone cuts in in front of you? Even though it would often be sufficient to just ease off the accelerator a bit, mine slams on the brake and slows down to where the car behind has to brake hard was well. Then it takes for ever to get back up to speed , opening an even wider gap in front, thus inviting more cars to cut in, etc. In those situations I tend to just turn it off. It didn't used to do that before the last iteration.

I'd like to mention that the AP1 system is inconsistent in this behavior. I have experienced three basic cases: (1) Your case, in which the car decelerates more abruptly than you or I would think a human driver would; (2) the case in which a car slides into the lane in front of you and the car makes very little immediate adjustment and lets the space between cars gradually return to normal (which is more like what I would do, and I think is the most desirable behavior); and (3) cases in which something pulls in front of me, too close for comfort, and the car does nothing and my intervention is (apparently) necessary to avoid a collision.

This last case happened to my spouse and me twice on a recent trip down I-95 from Boston to Philadelphia. Both incidents involved tractor trailers, and I think both were in New York or New jersey fairly close to NYC where traffic is heavy and drivers are fairly aggressive at changing lanes. The trucks signaled, and if I had been driving without AP1 I would have decelerated to allow them in. (In other words, i am not blaming the truck drivers for inappropriate behavior.) As it was, the car did not seem to notice the truck or to perceive it as an obstacle, given that the car did not decelerate, and I chose to manually apply brakes, fairly hard both times. Both times, my spouse was quite alarmed. Needless to say, I do not like to alarm my spouse or other passengers! The second time it happened, I noted that the truck was a flatbed, with a fairly small load. That is, the trailer may not have "looked" like a solid object to the camera and/or radar. So although i do not recall the details of these two incidents that well now, I have wondered since then if perhaps a flatbed trailer without a visually large object near the back end might be a poor target for the AP1 system, such that it does not really "see" the trailer.

When incidents like this occur, I do not always mentally record the circumstances very accurately, as I am busy deciding what to do, and doing it. And thus I lose awareness of what might have appeared on the instrument panel display to give me a clue about what the situation looked like to the AP1 system.

 

[Cosmacelf]

What continues to puzzle me is how the instrument cluster shows the lines and that the car is moving closer to one than the other, or even over the line, and yet the system can't keep the car in the middle even though it "knows" the car is off-center. (Plus I have a memory that it was better in the past, but not certain my recollection is correct.)

That's happened to me too, and I too think it used to be better.

My theory as to why this happens is that the actual driving algorithm (which is 100% Tesla code) is now running out of resources (i.e. The CPU is being asked to do too much at once), whereas it did not in the past. Why? Because the code has been larded up with new features, a new UI, etc. I am hoping the long, long awaited new release with the new operating system update will help.

Otherwise I was glad to have it. Five+ hours of driving is a lot for me, and I feel safer having the AP keeping an eye on things for me.

I agree, even if it is slowly degrading. It is still useful and I'm glad to have it too.

 

[smilepak]

* When the car in front kept me from going at preset speed and then it changed lanes opening up my lane, the car used to bolt forward like a jackrabbit (sort of). Not anymore. The acceleration is smooth, starting out slow and building momentum, rather than jerk forward.

Oh yes! This was the #1 issue I had. I hate it bolting forward, make me nervous to use during situation like heavy traffic and the car in front of moves to a different, it would bolt really fast.