Autopilot 7.0: When to be most on your toes?

[Denarius]

Good thread.

I do think lane-biasing has a lot of potential. For example:

- detect vehicles that are laterally near you ... lane bias away from them.
- detect large semis or trailers that have high sway ... lane bias away from them.
- in a curve, lane bias should be towards inside of the curve as you enter, then fan out to center. Except when:
..........in a curve to the right and you're in the leftmost lane with room, lane bias to the left throughout the curve, especially with a car to the right
- when in the rightmost lane, deference should be given to the left lane marking, not the right one due to exits
- in overtake lanes, the car should know how to at least follow the slower vehicle in front, or, choose the rightmost lane. Right now, it just freaks.
- when in the leftmost lane and there is sonar detectable k-rail, the car should still bias left instead of right so it stays away from traffic. Barriers don't jump out at you, but other drivers do. Even centered is better than the current right bias. Exception when the barrier is closer than say, 4 feet.

I did extensive testing (4 hours of driving) last night on a two lane highway with windy roads. My Model S tended to hug the left line through the curves, which was very bad, as cars coming the opposite direction when entering to fast tend to go wide, coming close as well. 8-10 times I had to override autopilot to hug the right side of the lane to prevent a collision or at the very least a close call.

 

[bmah]

Good thread.

I do think lane-biasing has a lot of potential. For example:

- detect vehicles that are laterally near you ... lane bias away from them.
- detect large semis or trailers that have high sway ... lane bias away from them.
- in a curve, lane bias should be towards inside of the curve as you enter, then fan out to center. Except when:
..........in a curve to the right and you're in the leftmost lane with room, lane bias to the left throughout the curve, especially with a car to the right
- when in the rightmost lane, deference should be given to the left lane marking, not the right one due to exits
- in overtake lanes, the car should know how to at least follow the slower vehicle in front, or, choose the rightmost lane. Right now, it just freaks.
- when in the leftmost lane and there is sonar detectable k-rail, the car should still bias left instead of right so it stays away from traffic. Barriers don't jump out at you, but other drivers do. Even centered is better than the current right bias. Exception when the barrier is closer than say, 4 feet.

Indeed. I'd be much more comfortable with AP if only the first of these was implemented.

Some of these heuristics need to be flipped left-for-right in countries where people drive on the left side of the road.

 

[S4WRXTTCS]

What I'd like to see at a minimum is

Lane bias away from large/close vehicles
Dampening of the auto-steer. It feels like I'm with some driver who is looking immediately front and not down the road. I've been testing it with a few fingers on the steering wheel and there is way too much steering input going on.
Better handling of exits on the left and right lane. Default to not exiting.

In my experience with it performed great on I5 between Everett, WA and Shoreline, WA. I imagine this stretch was extensively tested by Tesla.

It didn't handle 405 all that well between Bothell, WA and Bellevue, WA in the left lane (next to the HOV lane(s)). The issue was the HOV has sections where you enter it and that confuses the autosteering. It performed much better in the middle lane.

 

[MikeBur]

When SOC is at 100%

One thing related more to TACC, though happened when testing Autosteer first, is that the car doesn't maintain set speed when going downhill when fully (100%) charged with TACC engaged. Does work as expected when not fully charged.

First discovered this Thursday after charging for long drive then setting TACC speed limit, and watching tesla not brake at all going downhill. I understand regen is not available, though was very surprised brakes aren't applied. It was most disconcerting and unexpected, you just don't expect to be doubling your TACC speed when engaged.

hadnt noticed this prior to 7, though maybe it's in 6.2, just coincidence I was trying this... Is this normal?

 

[simonog]

Indeed. I'd be much more comfortable with AP if only the first of these was implemented.

Some of these heuristics need to be flipped left-for-right in countries where people drive on the left side of the road.

The other challenge is that in Europe (including the UK) cars should keep to the inside lane at all times except when overtaking slower traffic. Travelling in an outer lane with nothing on say a middle lane can be penalised.

it makes for more disciplined driving (overtaking only on the outside for instance) but is a challenge for any auto-positioning as the car cannot choose any empty lane.

 

[stevezzzz]

Here's another situation to beware of. On I-80 Westbound from Reno to Auburn, Ca., Exit 139 provides an unusually tough solution for autosteer. Notice that two sets of white lines exist: the regular road lines and a thinnner set of white lines that leads the driver towards the turnoff, which is a hairpin turn and cannot be taken at highway speed. My Tesla detected that something was wrong, handed control back over to me, and I made a small correction to the left, back into the highway lane and away from the turnoff. This is precisely the type of setup that needs either attention by Caltrans to get rid of that second set of white lines or some type of exception transmitted to the Tesla through a Tesla database, to provide the right solution to the problem of multiple white lines on the highway.

Notice one of many thinner white lines in a row, diverging to the right from the highway lane marker

In this photo, you can see additional white lines trying to lead the Tesla right towards the hairpin turnoff. Fortunately, the Tesla AutoSteer had already turned control back over to the driver at this point and the driver was ready to act if AutoSteer didn't disconnect

To me, the 'additional, thinner white lines' look like leftover markings from an earlier alignment of the lanes (note that the lines continue beyond the exit), lines not thoroughly removed after the newer markings were added. Worn, ambiguous highway markings were cited by Elon a few months ago as a reason Autosteer was not ready for release as quickly as anticipated. I think we're all coming to realize he was not just blowing smoke.

 

[kirkbauer]

The other challenge is that in Europe (including the UK) cars should keep to the inside lane at all times except when overtaking slower traffic. Travelling in an outer lane with nothing on say a middle lane can be penalised.

it makes for more disciplined driving (overtaking only on the outside for instance) but is a challenge for any auto-positioning as the car cannot choose any empty lane.

They are talking about position within a single lane, not which lane to drive in, the driver chooses that.

 

[Khatsalano]

This may be stylistic, but auto-lane change is too aggressive in the line it chooses. I would prefer the angle of entry to be maybe -20%. Here's why. Right now, it feels like a lane change is a Z instead of an S curve. After the lane change, the recovery steer is just so aggressive it feels cheap or badly engineered to have that much lateral deceleration for a simple maneuver. It's the difference between touching down nicely on the runway (you know that feeling, especially as a passenger!) vs. impacting the tarmac hard because the pilot judged the flare wrong. I would strongly prefer a gentler lane change angle with subsequent gentler recovery steer.

- K

 

[LetsGoFast]

One thing related more to TACC, though happened when testing Autosteer first, is that the car doesn't maintain set speed when going downhill when fully (100%) charged with TACC engaged. Does work as expected when not fully charged.

First discovered this Thursday after charging for long drive then setting TACC speed limit, and watching tesla not brake at all going downhill. I understand regen is not available, though was very surprised brakes aren't applied. It was most disconcerting and unexpected, you just don't expect to be doubling your TACC speed when engaged.

hadnt noticed this prior to 7, though maybe it's in 6.2, just coincidence I was trying this... Is this normal?

If your charge limit is such that regen is disabled, I think it is normal. It appears that the speed model used by the car assumes regen and it doesn't accurately predict your speed when regen is off.

This may be stylistic, but auto-lane change is too aggressive in the line it chooses. I would prefer the angle of entry to be maybe -20%.

Interesting. I think it is sylistic, because I think many people take way too long to execute a lane change and think Tesla hits it pretty well on a straight road. If you perform a lane change on a curve, I do think it sometimes steers a bit too abruptly on the recovery turn. There are a number of stylistic tweaks I could imagine. For instance, I'd prefer to shade my lane position depending on the circumstances. I prefer to stay a bit left in the leftmost lane, giving center lane traffic a wider margin of error (especially when passing oversize vehicles). On a two lane highway, I'd prefer to stay further right to give oncoming traffic the widest margin. I'm noticing a lot of subtle things I do as a driver that autosteer can't easily replicate. In the center lane, I'd try to stay dead center unless either of the other two lanes has a temporary barrier, then I'd try to stay a bit away from that lane and so on. Some people prefer very long, gentle lane changes and others prefer to minimize the time spend astride the lane markers.

 

[Papafox]

I did extensive testing (4 hours of driving) last night on a two lane highway with windy roads. My Model S tended to hug the left line through the curves, which was very bad, as cars coming the opposite direction when entering to fast tend to go wide, coming close as well. 8-10 times I had to override autopilot to hug the right side of the lane to prevent a collision or at the very least a close call.

In the agreement each of us okayed in order to accept the beta use of autosteering, Tesla clearly insists that a center divider be present between you and the oncoming traffic. Here's the language:

"Autosteer is for use on highways that have a center divider and clear lane markings, or when there is a car directly ahead to follow. It should not be used on other kinds of roads or where the highways has very sharp turns or lane markings that are absent, faded, or ambiguous. Similar to the autopilot function in airplanes, you need to maintain control and responsibility for your vehicle while enjoying the convenience of Autosteer.

Do you want to enable Autosteer while it is in Beta?"

I think what you've discovered is that Tesla has good reasons for putting out that restriction on the autosteer at its current level of development. No doubt the product will improve before Tesla sanctions autosteer without a center divider. I am adding this center divider issue to the initial post of this thread.

 

[mspohr]

In the agreement each of us okayed in order to accept the beta use of autosteering, Tesla clearly insists that a center divider be present between you and the oncoming traffic. Here's the language:

"Autosteer is for use on highways that have a center divider and clear lane markings, or when there is a car directly ahead to follow. It should not be used on other kinds of roads or where the highways has very sharp turns or lane markings that are absent, faded, or ambiguous. Similar to the autopilot function in airplanes, you need to maintain control and responsibility for your vehicle while enjoying the convenience of Autosteer.

Interesting quote from the autosteer beta agreement.
I don't see anything similar for the general release. The part about "use on highways that have a center divider" seems to be missing from the general release of the software.
I don't know if they feel more confident about the software and now feel that this warning is not necessary.

 

[Papafox]

Interesting quote from the autosteer beta agreement.
I don't see anything similar for the general release. The part about "use on highways that have a center divider" seems to be missing from the general release of the software.
I don't know if they feel more confident about the software and now feel that this warning is not necessary.

Mspohr, you make a good point that the center divider language should be added to the release notes, as well. From reports I've seen, autosteer without a center divider is clearly not ready for prime time yet. Since drivers need to acknowledge the one paragraph statement from Tesla that addresses the need for a center divider as a condition for beta testing autosteer, I feel confident that the acknowledgement language is controlling.

 

[lg_golfs]

Did not see this listed. Apologies if I missed it.

Southern California HOV lanes. Its a double yellow which changes to dashed white when you are allowed to enter and exit the lane. My car wanted to exit most of the times the white dashes were transitioning back to the double yellow.

 

[Papafox]

Did not see this listed. Apologies if I missed it.

Southern California HOV lanes. Its a double yellow which changes to dashed white when you are allowed to enter and exit the lane. My car wanted to exit most of the times the white dashes were transitioning back to the double yellow.

lg_golfs,
I haven't heard of that challenge to autosteer yet. I'll include it on the first post. Thanks.

 

[MikeBur]

If your charge limit is such that regen is disabled, I think it is normal. It appears that the speed model used by the car assumes regen and it doesn't accurately predict your speed when regen is off.

Thanks.

I don't believe it was this bad previously; though range charging is a rare occurrence a heavy car going down a steep hill rapidly increases speed is unexpected behaviour for me. Still I've said my piece

 

[MikeBur]

Bad scare this morning

Had a bad scare this morning. If you saw a Model S do an impromptu sliding maneuver, in the rain, at 60mph, this morning on the I90 east bound near Seattle, that was me...

Not the fault of autopilot. A brand new (paper sticker) Mercedes 300 swerved into my lane and would have caused an accident if I had not been paying attention. Again, no fault of autopilot as the driver weaves sharply into my lane without indication or blind spot check. By the time the collision alert chimed I was already swerving into the hov lane to avoid. You know things are going to be fun when time slows and you purposely have to try to not react harshly - no time to curse or gesticulate - one of those moments...

Anywho, what does this have to do with autopilot? Well, autopilot was on - my hands actually resting on the wheel - when I had to swerve. I ended up over-correcting as the extra 1/10th of a second to overcome the "reluctance" of the wheel to cede control caused me to turn harder than I would normally have done. Then, when the lane steering disabled, the steering wheel became a lot more compliant and I was turning a lot more than I anticipated. I did slide a little and managed to get the car under control quickly, though I thank goodness that I used to rally drive many years ago. I had broke traction and at that speed in heavy traffic I was extremely grateful that I remembered some of my rainy Tarmac stages and didn't spin out.

Think it's worthwhile contacting TM and see if they want to pull and analyze the logs?

Btw I resisted the temptation to road rage the Mercedes; rather I continued behind them (on Autopilot) and watched them do the same thing to a motorbike. This took the proverbial biscuit, so I continued to follow them off the freeway, take some photos of the car and paper sticker, then passed this on to the authorities with the intimation they were likely drunk.

 

[Burt Reynolds]

MikeBur, that is crazy! Thank goodness for your attentiveness, quick action, and WRC skills!

I've had the same thought cross my mind when the car unexpectedly veers onto an exit ramp. As an attentive driver, you rapidly observe what is happening and you immediately take corrective measures. But your hand is met with a surprisingly reluctant wheel and a degree of force that is far in excess of any steering input that the Tesla has ever requires. This certainly delays the proper corrective steering input by a half second, and then throws off the rest of your corrective inputs when the stiff steering goes suddenly compliant. Perhaps the "I'm taking the wheel" programming needs to soften the effort necessary to take control. It's a bit of a wrestling match right now!

 

[jeff_adams]

Btw I resisted the temptation to road rage the Mercedes; rather I continued behind them (on Autopilot) and watched them do the same thing to a motorbike. This took the proverbial biscuit, so I continued to follow them off the freeway, take some photos of the car and paper sticker, then passed this on to the authorities with the intimation they were likely drunk.

You know, I think it might be a good idea if Tesla were to team up with a dash cam manufacturer and make a unit that could do a streaming upload to Highway Patrol or local police. Instead of road rage, just send in the evidence of reckless driving and let the authorities deal with them. With Geo location capabilities, the police should know exactly where the bad guys are at.

 

[DuluthDad]

Interesting comments about auto steer having problems transitioning from solid to dashed lines. In Atlanta, driving down I85 from Duluth to the city, the major highway has such a set up as you move thru a set of HOV/Toll lanes. I took this route to the airport Monday morning (daylight) and TACC and Autosteer performed well. It didn't attempt to "exit" when the lane line transitioned from solid white double to dashed white double lines. The only trouble AP had was navigating a few sharp left turns at the I85 and GA400 intersection and the I85 and I75 merge. Then back around to the left and right at the Grady Curve. For the majority of the 30 mile drive, AP was engaged and did about as well as I could expect. The biggest problem TACC has is seeing far enough ahead to slow the car down more gently than it does. Visual line of sight can see a standstill about a half mile away, TACC waits for about 1/10th of a mile at best and brakes hard. Looking forward to continued improvements - but this owner is very satisfied with the capability of this 'beta' software. Looking forward to watching it learn as I make this same drive every week.

 

[Soolim]

As an attentive driver, you rapidly observe what is happening and you immediately take corrective measures. But your hand is met with a surprisingly reluctant wheel and a degree of force that is far in excess of any steering input that the Tesla has ever requires. This certainly delays the proper corrective steering input by a half second, and then throws off the rest of your corrective inputs when the stiff steering goes suddenly compliant. Perhaps the "I'm taking the wheel" programming needs to soften the effort necessary to take control. It's a bit of a wrestling match right now!

I set my steering to sports, so quite used to the "more effort" type of steering. Hence the transition from AP to driver is not that bad for me. If Tesla were to set the torque sensing lower, it may discourage people to comply with the hands on wheel approach, due to too early AP to driver switch.