Life with Autopilot: 1500 miles trip report

[jbcarioca]

Today, 12:16 AM

#16​

jbcarioca

Member

Join DateFeb 2015LocationRio de Janeiro, Brazil and Coral Gables, FLPosts490

I like it too. I drove 130 miles with it today through a long construction area on a crowded Interstate highway (I71 approaching Cincinatti Ohio) and it handled it all better than I would have, especially because of all the heavy close trucks on one side, construction barrier on the other. Frankly I had no expectations it could handle that. Then I used it in a 25 mph zone full of intersections turn offs etc. I expected it to fail completely. It did better than many drivers, constantly switching from one line tracking to another. It had no problem with tut offs so long as the left line continued. Crossing intersections it darted about a bit but never left the proper trajectory. Candidly, I'm amazed! I know it will not fo everything so well.

I'll have a solid database. I have just begun a 1300 mile trip. The update completed, then I immediately began my trip.​

- - - Updated - - -

The previous post I made immediately after I installed the update. I was actually sitting in the car as it downloaded because I was late to begin my trip.

i will make daily reports of the experiences. From this morning I am certain future reports will not be so giddy as the first one.

For context: I am a pilot, Airline Transport Pilot, five type ratings, and have had experience testing new autopilots, including two certification tests. I am not an engineer but I have a modestly technical background. From time to time I will probably be comparing the Tesla AP to Aircraft AP's. Clearing the automotive problems are far harder to solve.

When I complete my 300 miles today I will make another post.

From my 200 miles so far I think this is rightly called a beta, but within those implications I think this is a magnificent accomshment.

1. The more unambiguous data this one has the better it performs. Unlike human drivers, dense traffic with clear markings and good contrast make the Tesla happy. As a human the dense traffic makes me nervous. Input to all twelve sensors plus camera and radar make Tesla very, very confident.
2. Open country, wide lanes, little traffic and all is OK, but Tesla hunts a bit.
3. If you're in the right lane in a tight urban area with tights turn offs, no problem. If you're on a rural broad Interstate and in the right lane,Tesla will try to turn off. Ditto for splits in the highway.

more later...

 

[Nevek]

Sounds great and I look forward to you posts on the trip. Your flight experience will make it particularly interesting.

I did about 200 miles with autopilot yesterday and my experience matches yours. I did find that on tight turns on secondary roads (not freeway but well marked rural highways) the car tended to "bounce" from one side of the lane to the other.

 

[ilovemycoffee]

Looking forward to your posts! Your posts at Flyertalk are very informative, and your travel experience will be very valuable.

 

[jbcarioca]

Today ended out being 350 miles. I have now used the AP in every type of road, in the dark, with glaring sun, and in construction zones, city streets and switching from one type of road to another.

Definitively there are things the AP cannot do. So far I have three major ones:
1. On wide roads the AP cannot handle being in the exit lane but not taking the exit. Soon after the exit side line departs for an exit the AP will try to exit also. Oddly, in city streets and other tight turns it does not try to leave. We'll see in subsequent days if that holds true.
2. You CANNOT rely on the AP if you are in a lane that merges with another. The moment the right and left lines merge the poor AP has a fainting spell!
3. Similarly NEVER expect the AP to understand if you are in a split of a superhighway. The poor AP will faint there also. Make sure you are in a lane that must go where you want to go.

Now for the unexpected good things:
1. The AP works quite well in city traffic. I know we are told it does not...but it does.
2. It takes a few hours using the AP to understand that it has tolerances tighter than ours. Thus, faced with high density traffic it is better than I am. So far I am slowing and taking control because I doubt I could handle the situation were it to fail. Lesson from aircraft: machines do it better, but people must cope when machines fail. Thus, don't allow the machine to do something you could not do well yourself.
3. Engaging the AP from a full stop does work, and works well. I used that feature several times today. There's a learning curve, and I expect mine will be steep on this issue.
4. If you enjoy drafting, the AP will be much nicer than was TACC alone. Another P85D driver I met twice today reported better energy efficiency in drafting than he had before. I concur with him.

Net: my two day view suggests using 7.0 is a complex task so studying and practice are important. In aircraft terms, we probably need a type rating to use 7.0.

Finally everything I liked that seemed to be missing is somewhere. I just must practice, read and search. I love this, but without doubt it is not "get in and go". It reminds me of an early BMW 750 iDrive, but with the intolerance for misunderstanding that goes with a Learjet 35. Translating, that means it is glorious, but must be learned.

 

[AudubonB]

Today ended out being 350 miles. I have now used the AP in every type of road, in the dark, with glaring sun, and in construction zones, city streets and switching from one type of road to another.

Definitively there are things the AP cannot do. So far I have three major ones:
1. On wide roads the AP cannot handle being in the exit lane but not taking the exit. Soon after the exit side line departs for an exit the AP will try to exit also. Oddly, in city streets and other tight turns it does not try to leave. We'll see in subsequent days if that holds true.
2. You CANNOT rely on the AP if you are in a lane that merges with another. The moment the right and left lines merge the poor AP has a fainting spell!
3. Similarly NEVER expect the AP to understand if you are in a split of a superhighway. The poor AP will faint there also. Make sure you are in a lane that must go where you want to go.

Now for the unexpected good things:
1. The AP works quite well in city traffic. I know we are told it does not...but it does.
2. It takes a few hours using the AP to understand that it has tolerances tighter than ours. Thus, faced with high density traffic it is better than I am. So far I am slowing and taking control because I doubt I could handle the situation were it to fail. Lesson from aircraft: machines do it better, but people must cope when machines fail. Thus, don't allow the machine to do something you could not do well yourself.
3. Engaging the AP from a full stop does work, and works well. I used that feature several times today. There's a learning curve, and I expect mine will be steep on this issue.

4. If you empty drafting, the AP will be much nicer than was TACC alone. Another P85D driver I met twice today reported better energy efficiency in drafting than he had before. I concur with him.

Net: my two day view suggests using 7.0 is a complex task so studying and practice are important. In aircraft terms, we probably need a type rating to use 7.0.

Finally everything I liked that seemed to be missing is somewhere. I just must practice, read and search. I love this, but without doubt it is not "get in and go". It reminds me of an early BMW 750 iDrive, but with the intolerance for misunderstanding that goes with a Learjet 35. Translating, that means it is glorious, but must be learned.

Empty" = "enjoy", perhaps? I'm really scratching my head on this one." Vacilando, vazia vacilando....

 

[eye.surgeon]

Thanks for the interesting observations. As an aside, please don't draft. It's dangerous and the savings is literally trivial.

 

[bollar]

Thanks for the interesting observations. As an aside, please don't draft. It's dangerous and the savings is literally trivial.

If autopilot has him following a truck at what AP considers to be a safe distance, I wonder if it's truly drafting? Interesting.

 

[jbcarioca]

Thanks for the interesting observations. As an aside, please don't draft. It's dangerous and the savings is literally trivial.

Drafting as Traditionally done is dangerous, no doubt. Drafting at a 2 second interval exceeds tradional following recommendations. Adaptive cruise control also offers improved control. The tradional did not allow adequate time for response. Technology helped.

- - - Updated - - -

Empty" = "enjoy", perhaps? I'm really scratching my head on this one." Vacilando, vazia vacilando....

Yes, enjoy. My typo. Corrected.

 

[eye.surgeon]

The danger I am referring to, at least in part, is the road rage you will incur. Adaptive cruise control or not.

 

[BriansTesla]

Thank you very much for your insights. A pilot's viewpoint is very helpful.

 

[jbcarioca]

Day three:
I drove 250 miles today, primarily from Knoxville to Charlotte via Asheville. Thus I drove in mountainous terrain assuming you consider the Smokeys to count.:wink:
Several new things happened today:
1. Fog seems to have little effect on AP;
2. early and late bright sun, hitting on an angle, tends to blind the AP sensors;
3. Midday bright sun has no effect at all;
4. Losing a center lane marking usually does not cause the auto-steering to fail, but...
5. Losing an outside lane marking making the AP panic and quit;
6. even with the odd line disappearance if there is traffic on both sides the auto-steering continues happily;
7. The auto-steer absolutely cannot reconcile merging lane marking as in; lane merges, entry and exits, freeway splits. If marking continue on both sides of an exit the auto-steer wobbles a little but does not fail.
Later today In began to experiment a little bit and discovered some new tactics:
7. Coming up on an unmarked (with street stripes) exit- holding the steering wheel firmly but with pressure, NOT actual movement, usually keep everything normal;
8. On all curves, but most pronounced on long sweeping curves, auto-steer makes a continuing series of small turns, straightens and turns again, producing a jerky turning process.
8. Upon that discovery I tried stabilizing the auto-steer with light pressure when it vacillated on curves. That worked.
While trying to understand some of the behavior I reflected on the difference between "rate-based' and 'altitude-based' autopilots in terms of smoothness. These days those distinctions are not so important as they were for such GA models as early S-Tec's and the Bendix-King KFC-150 etc. The rate-based were smooth but sloppy, so could remain connected no matter what turbulence was encountered while the altitude-based were precise but could easily overstress the airframe so could not be used in turbulence. Obviously the technologies needed to evolve to have both advantages with neither disadvantage.
Similarly auto-steer function is now very conservative. It resigns whenever there are ambiguous signals even if the probabilities are clear. It remains slightly jerky in directional change, but manages some quite demanding turns. I'm thinking these characteristics will improve rapidly as 50,000 drivers contribute data. That is a lot of beta-testing.

Finally:
9. A others have commented, in traffic there si no longer the 10 seconds or so then enter hold, I have yet to experience a hold.
10. Starting, stopping and slowing from a long distance between cars at an intersection is far, far smoother and less dramatic than it was.
11. Engaging auto-steer and TACC from a stop when following another car works and works well.
12. No question all, v 7.0 works optimally in fairly heavy, stop and go traffic. there is is nearly flawless.

It may be simply that my learning curve is continuing, but the entirety of v7.0, seems to be learning. Elon intimated a continuous learning process distinct from software releases, although I did not take that seriously. is it really improving? I am not sure. I cannot distinguish between v7.0 doing better and me learning how to use it better, at lest now on day 3.

The displays are another subject but I will say I miss a few things, selectable data on trips and current trip energy usage, instant readout of current energy usage etc. I expect and hope we'll soon have more customization so we can choose what we want to have from a broader menu.

I also have learned to like the toy cars, especially the blind spot warnings and the visual show of vehicles ahead.

Tomorrow I will have only local driving around Charlotte so I'll have less to report.

On an unrelated note: I always try to stay plugged in in order to reduce stress on the 12V system. I have three times on this trip found that hotels often have level one outlets around the property but are often unaware of them. I have had three hotels quite happy to allow me to plug into them. I find that I consistently gain 3-4 mph from them, which can be significant if you'll be staying overnight. Above all you'll have no stress on the 12V system and will ahem no vampire losses. Twice I have charged at a Supercharger 20-30 miles distant and recovered my loss with the level one overnight. I'm grateful for that and I carry a 25 foot/12 gauge extension cord to ensure I can reach those plugs.

 

[CharlesMW]

Good Analysis - I found very much the same.
Love Autopilot in stop / start heavy traffic into the city.

The new "toy car" display would be amazing with a heads-up display - Hopefully in a future version of the S ..... please

 

[cgiGuy]

The danger I am referring to, at least in part, is the road rage you will incur. Adaptive cruise control or not.

From the truck driver? They're quite used to it. Just let them know you're back there.

 

[jbcarioca]

This morning i have been thinking of two largely unrelated points:

1. Wherever I need something, from a good hand car wash to finding overnight charging people are genuinely excited to help a Tesla driver. In addition, it seems every Prius driver is excited to talk Tesla. This trip is showing me that such findings are independent of how common Tesla's are, although i rather doubt such enthusiasm is common in, say, the San Francisco Bay area.

2. Nearly everyone seems to diss level 1 charging. However, I am finding quite good experience with using level 1 on this trip. Here is why; if, as in Cincinnati, Ohio, the Supercharger is located in one area and your trip and hotel in another further along your route, you have the choice to backtrack to top up, maybe range charging, or find a place to get 30-50 miles added overnight. Level 1 does that nicely. Three times in a row I have found no Destination charger nor PlugShare location near my route. I call my hotel and ask if they have EV charging and they do not. I ask if they have an outside 120v outlet and the situation changes. Again three times in a row I have been put on hold while a maintenance person goes to check, invariably finding an outlet. They've been quite happy for me to use it. I bought a 25ft/12 gauge extension cord just to reach these places. So far I have consistently gained about 3mph, enough to eliminate the need for range charging and simultaneously reducing the stress on my 12v system by remaining connected. So, count me as a fan for level 1. Hopeless for most situations, ideal for this one.

 

[lolachampcar]

jbc,
Being from the certified world, you have likely not been exposed to AP control loop values (or PID values from an engineering view). On the experimental stuff, we get to play with the proportional, integrative and derivative gains to tweak how the AP functions. All planes are different and the AP trimming can be entertaining.

Anyway, my take on the lane darting in corners are that Mobile Eye is allowing too much error before correcting and that correction is, likewise, too much. Simply reducing the time constant such that sample/corrections occur at a higher rate may fix this as the errors and corrections should be proportionally smaller.

As for the AP's ability to deal with exits, merging and lane splits, there is no navigation tie to the system thus it is simply following lane markers. Perhaps a bias towards the non-exit side lane marker would protect against interstate exit but then I have had my car aggressively dart into the center lane of a three lane road when an upcoming center turn lane is clearly denoted by the lane markers. Perhaps a bias towards straight when one lane marker departs would address this as well. As for mergers and splits, without a navigation tie I have no idea how the AP could know what to do.

Lastly, I share your enthusiasm. For a first pass, Tesla has knocked it out of the park.

On the hardware side, we need to realize that almost everyone is using Mobile Eye. Every other Mobile Eye customer with steering and braking integration could allow what Tesla has allowed. Their size prevents them from taking such bold and risky steps. They will likely learn from what Tesla is doing and come close on their heals as a safer second/third/forth.

 

[MsElectric]

If autopilot has him following a truck at what AP considers to be a safe distance, I wonder if it's truly drafting? Interesting.

The problem with following a large truck closely is you use all situational awareness as a lot of your vision is blocked. As a rule I never drive on a highway with large vehicles in front of me. I want to know what is coming up ahead and if something unexpected happens, I want to have an escape route that i can figure out really quickly.

Another reason are the horrible tractor trailer bumpers in the US. I heard in Canada, they are supposed to be reinforced and at the same height as a car bumper. In the US, they are sometimes flimsy and even if TACC does stop in time, if another car rear ends you, you could get forced under a tractor trailer.

 

[e-FTW]

Lesson from aircraft: machines do it better, but people must cope when machines fail. Thus, don't allow the machine to do something you could not do well yourself.

Word.

 

[jbcarioca]

jbc,
Being from the certified world, you have likely not been exposed to AP control loop values (or PID values from an engineering view). On the experimental stuff, we get to play with the proportional, integrative and derivative gains to tweak how the AP functions. All planes are different and the AP trimming can be entertaining.

Anyway, my take on the lane darting in corners are that Mobile Eye is allowing too much error before correcting and that correction is, likewise, too much. Simply reducing the time constant such that sample/corrections occur at a higher rate may fix this as the errors and corrections should be proportionally smaller.

...

On the hardware side, we need to realize that almost everyone is using Mobile Eye. Every other Mobile Eye customer with steering and braking integration could allow what Tesla has allowed. Their size prevents them from taking such bold and risky steps. They will likely learn from what Tesla is doing and come close on their heals as a safer second/third/forth.

Good points. I have been involved with two autopilot STC's, one for a single engine airplane, the other for a Learjet. The PID was easy on the slow, little guy, but was very hard on the Lear, which was done on a 35A, one of the most challenging non-military aircraft ( little known fact: the Lear was derived from a failed candidate for a Swiss Air Force fighter competition. The early Lears, the model 23 crashed with such frequency that the modern type rating evolved from that experience). I digress, but the Model S is itself a different beast.

I agree with you that increasing sampling rate might improve cornering smoothness. Evidence: slower corners are smoother than faster ones. However, I think that the longer range sensitivities are a weakness that probably needs to be addressed with better radar sensitivity and improved longer range data interpretation. The radar limitations are quite hard to resolve because of longer range 'noise'. Still, the smoothness and stability, especially in corners do demand improved predictive ability, and that in part means sensor tuning ( including camera and radar in that definition).

Another issue, as I experienced yesterday driving around Charlotte, NC is that she cannot now deal with faded lane markers or non-existent ones, which abound in otherwise well-maintained roads there. Of course most larger cities have many problems like that, and worse.

As you imply, the upcoming incirporation of GPS, mapping and all the Beta data will be transformative. That will be amazing!

in the meantime I do worry that some of us fail to understand that we are driving a Beta version. The complaints illustrate that lack of understanding or, maybe more likely, the tactics of people who have shorted TSLA. For the record, I remain long on TSLA.

- - - Updated - - -

The problem with following a large truck closely is you use all situational awareness as a lot of your vision is blocked. As a rule I never drive on a highway with large vehicles in front of me. I want to know what is coming up ahead and if something unexpected happens, I want to have an escape route that i can figure out really quickly.

Another reason are the horrible tractor trailer bumpers in the US. I heard in Canada, they are supposed to be reinforced and at the same height as a car bumper. In the US, they are sometimes flimsy and even if TACC does stop in time, if another car rear ends you, you could get forced under a tractor trailer.

Maybe I have confused this issue by using the word "drafting". That word is NOT synonymous with "tailgating" as I am using it, at least.

using TACC set on 1, still provides safe distance, and allows the preceding vehicle to see the following one. Anything closer that that would be unsafe and imprudent, no doubt.

The aerodynamic benefit of reduced wind resistance is certainly lower when one follows safely, but collision repair and injury could outweigh the benefits:crying:

In any event following safely still can reduce energy usage by 10-20%, depending on weather conditions, including wind, precipitation and temperatures, as well as altitude changes. There are far too many variables for me to measure, although every vehicle manufacturer studies all these factors and more.

 

[SR22pilot]

Minor detail but 'altitude-based' autopilots should be 'attitude-based' autopilots.

 

[lolachampcar]

so they fly differently depending upon how much of a d*** the pilot is?????