17.17.4

[_jal_]

... It all stemmed from the idiot with the water bottle on the steering wheel trying to get around these warnings.

I'm going to go into work tomorrow and act all steamed. When someone asks me what's wrong I'm going to bark "I don't even remember, but some idiot put a water bottle on his steering wheel and that was the last straw!"

Honestly for our entire lives we've waited for someone to build a car that drives itself. We finally get the first whiffs of that but it comes with the caveat "No one has ever done this before. Please [it actually says 'please'] keep your hands on the wheel" and immediately human nature attempts to circumvent this simple and modest warning. How much further we'd go as a species if we'd Just Read the Instructions.

 

[chillaban]

I’m glad there’s at least some optimists on the forum. Seriously, just a few years ago if you let go of the steering wheel even on a fairly straight road you’d depart your lane in a matter of seconds.

And just 3 years ago and even today for non-Teslas, lane keeping systems just sorta nudge your steering and if you let go of the wheel they last for maybe 10 seconds or less before giving out.

Now Tesla has delivered two industry first systems available to customers today where you can find a type of road, with ap1 or AP2, where you can let go of the wheel for literally minutes and the car drives reasonably. Yes we would love to see it do more and handle more types of situations, but what we already have is pretty notable.

 

[oktane]

I’m glad there’s at least some optimists on the forum. Seriously, just a few years ago if you let go of the steering wheel even on a fairly straight road you’d depart your lane in a matter of seconds.

And just 3 years ago and even today for non-Teslas, lane keeping systems just sorta nudge your steering and if you let go of the wheel they last for maybe 10 seconds or less before giving out.

Now Tesla has delivered two industry first systems available to customers today where you can find a type of road, with ap1 or AP2, where you can let go of the wheel for literally minutes and the car drives reasonably. Yes we would love to see it do more and handle more types of situations, but what we already have is pretty notable.

I agree with this. To me it's both remarkable and insulting at the same time. I'd be more or less happy with the car if only Tesla had been honest to me.....

 

[Tangible]

even today for non-Teslas, lane keeping systems just sorta nudge your steering and if you let go of the wheel they last for maybe 10 seconds or less before giving out.

Last night, in the dark with lane markings ravaged by a winter of salt and scraping plows, my 2015 Acura MDX drove itself for 15 minutes on I-95 in moderately heavy traffic. My only input was a required touch of the steering wheel every two minutes or so. The car stayed centered in its lane and kept a consistent distance from the car ahead, at about 70 mph.

The only anomaly happened when a vehicle going much faster than me cut into my lane only about one car-length ahead of me. As a driver I would have ignored it (except for a mumbled expletive) because of our relative speed. The MDX braked sharply, and then had to accelerate back to cruising speed.

In stop and go traffic it is fully automated, except that if it comes to a full stop it insists on a touch of the Resume button. On local streets it's erratic, just like my Model S.

This is a two-year-old $50,000 car. Though it modestly calls its features Lane Keeping Assistance and Adaptive Cruise Control, it's pretty much the equal of AP2 EAP in its current state.

PS: When I began my drive in the MDX, which is my wife's regular ride, I put it in Drive by pushing down on the stalk to the right of the steering wheel, and the vehicle didn't move. However, the windshield wipers and washer turned on. After a second or two the Wetware system was reprogrammed, and all went well thereafter.

 

[oktane]

Last night, in the dark with lane markings ravaged by a winter of salt and scraping plows, my 2015 Acura MDX drove itself for 15 minutes on I-95 in moderately heavy traffic. My only input was a required touch of the steering wheel every two minutes or so. The car stayed centered in its lane and kept a consistent distance from the car ahead, at about 70 mph.

The only anomaly happened when a vehicle going much faster than me cut into my lane only about one car-length ahead of me. As a driver I would have ignored it (except for a mumbled expletive) because of our relative speed. The MDX braked sharply, and then had to accelerate back to cruising speed.

In stop and go traffic it is fully automated, except that if it comes to a full stop it insists on a touch of the Resume button. On local streets it's erratic, just like my Model S.

This is a two-year-old $50,000 car. Though it modestly calls its features Lane Keeping Assistance and Adaptive Cruise Control, it's pretty much the equal of AP2 EAP in its current state.

PS: When I began my drive in the MDX, which is my wife's regular ride, I put it in Drive by pushing down on the stalk to the right of the steering wheel, and the vehicle didn't move. However, the windshield wipers and washer turned on. After a second or two the Wetware system was reprogrammed, and all went well thereafter.

You mean my new enhanced autopilot full self driving $155,000 Tesla actually has autopilot functionality similar to a 2 year old $50,000 Acura??? And I'm supposed to love Tesla for this?

 

[Sawyer8888]

I still have 17.17.4 and I recently discovered that when at a complete stop, if I lightly press the brake and release, my break lights do not stay engaged. If on the other hand I push the break in further and then release, my breaks do stay engaged until I press the accelerator. This is all based on what I see in my display and while on inclines and declines.

This may have been a feature in a Tesla and many others long before I upgraded to a nicer car (I did not search the forum before writing this). Coming from a 9-yr-old Honda, this added benefit of newer technology and/or higher-end vehicles is quite pleasant.

...now back to autopilot discussions.

 

[BestRadar]

I still have 17.17.4 and I recently discovered that when at a complete stop, if I lightly press the brake and release, my break lights do not stay engaged. If on the other hand I push the break in further and then release, my breaks do stay engaged until I press the accelerator. This is all based on what I see in my display and while on inclines and declines.

This may have been a feature in a Tesla and many others long before I upgraded to a nicer car (I did not search the forum before writing this). Coming from a 9-yr-old Honda, this added benefit of newer technology and/or higher-end vehicles is quite pleasant.

...now back to autopilot discussions.

This is normal. Look for the break hold icon that will initiate when you first stop. If you tap the break pedal you are telling the break hold to turn off. This has been like this all along but you probably just never noticed.

 

[Aeg07046]

After putting about 2000 miles on a new MS with half on 17,17,4, I have one major complaint t about Autosteer I need the self-drive/autosteer experts on this board to comment on. On highways, the car is very good in figuring out where the lanes are. However, almost always, the car favors the right side of the lane and "knowingly" puts the car uncomfortably close to the right lane marker. I just don't understand this behaviour. The Tesla engineers have figured out the really hard thing which is correctly identifying the lanes but fail completely on the easiest thing: a nanosecond math calculation to determine the center of the lane and direct the car to it. It is almost that they don't trust their lane determination still. Is it possible that there is something wrong with my car or its calibration. I did see this mentioned elsewhere.

The second issue is that the "upgraded" HW2 sensors are useless at speed. It takes them much too long to recognize objects on the right or left of the car at speeds of over 50 mph for their use to matter. In other words, if the car is at 60 mph, and is veering into the car on the right lane, the sensors will not let it avoid a collision since by the time they recognize the car on the right lane, you would have already hit it. Again, am I the only one wit this condition?

 

[Aeg07046]

After putting about 2000 miles on a new MS with half on 17,17,4, I have one major complaint t about Autosteer I need the self-drive/autosteer experts on this board to comment on. On highways, the car is very good in figuring out where the lanes are. However, almost always, the car favors the right side of the lane and "knowingly" puts the car uncomfortably close to the right lane marker. I just don't understand this behaviour. The Tesla engineers have figured out the really hard thing which is correctly identifying the lanes but fail completely on the easiest thing: a nanosecond math calculation to determine the center of the lane and direct the car to it. It is almost that they don't trust their lane determination still. Is it possible that there is something wrong with my car or its calibration. I did see this mentioned elsewhere.

The second issue is that the "upgraded" HW2 sensors are useless at speed. It takes them much too long to recognize objects on the right or left of the car at speeds of over 50 mph for their use to matter. In other words, if the car is at 60 mph, and is veering into the car on the right lane, the sensors will not let it avoid a collision since by the time they recognize the car on the right lane, you would have already hit it. Again, am I the only one wit this condition?

 

[Sawyer8888]

After putting about 2000 miles on a new MS with half on 17,17,4, I have one major complaint t about Autosteer I need the self-drive/autosteer experts on this board to comment on. On highways, the car is very good in figuring out where the lanes are. However, almost always, the car favors the right side of the lane and "knowingly" puts the car uncomfortably close to the right lane marker. I just don't understand this behaviour. The Tesla engineers have figured out the really hard thing which is correctly identifying the lanes but fail completely on the easiest thing: a nanosecond math calculation to determine the center of the lane and direct the car to it. It is almost that they don't trust their lane determination still. Is it possible that there is something wrong with my car or its calibration. I did see this mentioned elsewhere.

The second issue is that the "upgraded" HW2 sensors are useless at speed. It takes them much too long to recognize objects on the right or left of the car at speeds of over 50 mph for their use to matter. In other words, if the car is at 60 mph, and is veering into the car on the right lane, the sensors will not let it avoid a collision since by the time they recognize the car on the right lane, you would have already hit it. Again, am I the only one wit this condition?

I believe we are all seeing the same thing on both counts. I find it quite annoying and a bit dangerous but I'm hopeful that Tesla will eventually get it right.

 

[oktane]

Accurate. Side ultrasound refresh too slowly to be useful.

After putting about 2000 miles on a new MS with half on 17,17,4, I have one major complaint t about Autosteer I need the self-drive/autosteer experts on this board to comment on. On highways, the car is very good in figuring out where the lanes are. However, almost always, the car favors the right side of the lane and "knowingly" puts the car uncomfortably close to the right lane marker. I just don't understand this behaviour. The Tesla engineers have figured out the really hard thing which is correctly identifying the lanes but fail completely on the easiest thing: a nanosecond math calculation to determine the center of the lane and direct the car to it. It is almost that they don't trust their lane determination still. Is it possible that there is something wrong with my car or its calibration. I did see this mentioned elsewhere.

The second issue is that the "upgraded" HW2 sensors are useless at speed. It takes them much too long to recognize objects on the right or left of the car at speeds of over 50 mph for their use to matter. In other words, if the car is at 60 mph, and is veering into the car on the right lane, the sensors will not let it avoid a collision since by the time they recognize the car on the right lane, you would have already hit it. Again, am I the only one wit this condition?

 

[Tangible]

almost always, the car favors the right side of the lane

Yesterday I took a two hour drive on the Mass Pike with AP2 engaged almost the whole time. The bias toward the right was obvious, and whenever I passed an 18-wheeler the car seemed to pull toward it, causing me to take over twice.

Here's an alternate hypothesis:

We all believe, without proof, that we drive in the center of our lane. What if it's not true, and our instinct is actually putting us left of center? Our driving position at the left edge of the car rather than in the center certainly creates some distortion.

Further, what if we unconsciously shy away from large trucks that we're passing, and see the Tesla's indifference to the intimidating presence as actual attraction?

I don't actually believe this is the case but I think it's worth exploring.

It would also be interesting to hear from U.K. owners to learn if the bias seems to be to the left edge there.

 

[chillaban]

My thoughts on "pulling to the right":

My S100D was delivered with an alignment issue where the car pulled to the right if I let go of the wheel. EPAS eventually compensates for it, but you can still see that when you are driving the car straight, the "T" logo on the wheel is clearly tilted.

In this state, AP2 would always return the wheel to neutral as soon as I engage, resulting in a pretty bad pull to the right. After they aligned my steering, this problem went away.

But AP2 drives dead center, and that's not what humans do when driving next to a truck. AP2 does not proactively avoid driving too close to a truck or other wide car in a nearby lane. It might reactively move over a little once you're already side-by-side if the ultrasonics see it. But most likely, the majority of complaints about the right bias is that we as humans are not comfortable with driving dead center.

 

[doubleohwhat]

What's the latest update for an AP1 S? I'm on 17.11.10

 

[Tangible]

But most likely, the majority of complaints about the right bias is that we as humans are not comfortable with driving dead center.

Ok, now that there are two of us it's a movement!

What we need now is an accomplice taking a video through the rear window of a front-running vehicle, comparing the actual lane position to the perception of the Tesla driver. Too bad TMC isn't a real club, where we could do this as a bar bet and then hit the road to try it. Like you I'm in the Bay Area, but it's Massachusetts Bay.

Still waiting for a Brit to chime in: Do they perceive left bias?

 

[chillaban]

Ok, now that there are two of us it's a movement!

What we need now is an accomplice taking a video through the rear window of a front-running vehicle, comparing the actual lane position to the perception of the Tesla driver. Too bad TMC isn't a real club, where we could do this as a bar bet and then hit the road to try it. Like you I'm in the Bay Area, but it's Massachusetts Bay.

Still waiting for a Brit to chime in: Do they perceive left bias?

As a tip, I have a second user profile with the mirrors tilted down a bit. That's the best way to compare right vs left lane market clearance.

 

[Aeg07046]

I don't think the "Right Bias" is a human perception issue. In my case, on the display where AP plots the lane lines, it shows the car on the right and very close to the right lane lines that it has detected. What it shows on the AP display, Right Bias, matches my human perception of Right Bias. So, it is a programming issue. Tesla Engineers decided to have a Right Bias knowingly. Either they had issues with lane detection in the past or they feel that when lane detection becomes unreliable at some point during the journey, it is better to have a bias, but this is a conscious decision on their part.

 

[Tangible]

Tesla Engineers decided to have a Right Bias knowingly.

I think you're more certain about this than the available evidence suggests. I don't think we know to what degree the dashboard display reflects the data upon which AP is making driving decisions. We've all seen the lines and the vehicles jumping around on the display —less so in 17.17.4 than in the past, but still very pronounced.

As always, a little transparency from Tesla would be welcome. Until that day comes, we really have no idea what's going on in the design labs.

 

[sunnyvale]

anyone got 17.18.50 ?

 

[J1mbo]

Ok, now that there are two of us it's a movement!

What we need now is an accomplice taking a video through the rear window of a front-running vehicle, comparing the actual lane position to the perception of the Tesla driver. Too bad TMC isn't a real club, where we could do this as a bar bet and then hit the road to try it. Like you I'm in the Bay Area, but it's Massachusetts Bay.

Still waiting for a Brit to chime in: Do they perceive left bias?

Nope. Before 17.17.4, when I thought the car was too close to the car on the left, I checked the mirrors and my car was clearly dead center.

It is a perception issue, amplified by the fact that the car on the left is rarely dead center themselves.