Autopilot disengagement, driving etc. out of main.

[ZsoZso]

I'm curious, for those of you on the latest no-confirm navigate on autopilot: what's your miles-per-intervention rate on it? I'm wondering what our baseline is, before going from HW2 to HW3.

Well to be honest, I am not a huge fan of navigate-on-autopilot. I have a 2 hour commute per work day (1 each direction), most of it (~90%) on highway where I use autopilot extensively. But, NoA, I do not like and mostly due to its choices of lanes. It does not want to use the HOV lane, which is dumb when the traffic is heavy (usual) on the main lanes. It also wants to got out to the slowest lane ~3km before the exit, which is again dumb as it gets behind slow trucks. Then it starts overtaking them, sometimes very close to the exit, yet another dumb thing.

While AP itself is getting better, even lane changes requested by indicator is quite good now (used to be finicky, undecided, taking too long), but the lane choice logic of NoA is far too primitive. I am sure it is more useful for driving routes that you do not know much and would need to rely on the navigation to know which interchange to take. However, for routine daily driving I can make much better lane choices based on experience. For this reason I rarely use NoA, once in a while after each sw update to see how they improved it, but so far I am not impressed so I stick to basic AP on daily use.

 

[Huskyf]

Well to be honest, I am not a huge fan of navigate-on-autopilot. I have a 2 hour commute per work day (1 each direction), most of it (~90%) on highway where I use autopilot extensively. But, NoA, I do not like and mostly due to its choices of lanes. It does not want to use the HOV lane, which is dumb when the traffic is heavy (usual) on the main lanes. It also wants to got out to the slowest lane ~3km before the exit, which is again dumb as it gets behind slow trucks. Then it starts overtaking them, sometimes very close to the exit, yet another dumb thing.

While AP itself is getting better, even lane changes requested by indicator is quite good now (used to be finicky, undecided, taking too long), but the lane choice logic of NoA is far too primitive. I am sure it is more useful for driving routes that you do not know much and would need to rely on the navigation to know which interchange to take. However, for routine daily driving I can make much better lane choices based on experience. For this reason I rarely use NoA, once in a while after each sw update to see how they improved it, but so far I am not impressed so I stick to basic AP on daily use.

I saw on youtube videos that NOA use HOV lane correctly but you have to activate it on the menu Navigation or Autopilot I'm not sure sorry.

 

[bdy0627]

Not correct, in traffic queues, EAP keeps pretty close to the car in front.

The only issue it has is that it stays in the centre of the lanes and if there's a motorbike coming through you'll need to disengage and move over to the side to give space - at least on European highways.

I wish the car would stay a bit closer to the one in front on the most aggressive setting for dense highway driving. There is too much room for drivers to cut in front simply because AP allows more space than human drivers, so it is the pathway of least resistance.

 

[intelligator]

... Has anyone sat back in their chair and thought about the practicality of FSD? You can't use it on major highways in gridlock traffic or in large cities, as it will try to maintain a good, safe distance from the car in front of it which means other drivers and taxi cabs will jump into any opening. You will be lucky to move. Going cross town in mid town Manhattan during daylight hours? Hahahahaha!!!!! ...

I live between philly and nyc and get your point, clearly. However, I personally welcome letting an assist system deal with the stress of aggressive merges, cut offs, etc. Maybe a cabby could become the needed lead vehicle? I can see where some drivers that try to close the gap create more of a risk for themselves, rear ending, etc. If traffic were at a stop or someone has to get somewhere faster than I, let it be. When I have the same need, I opt out of driving assist and take full control. Meanwhile and for the purpose of safely getting to my destination with reduced stress, I will continue to use NoA whenever possible.

 

[Dan Detweiler]

Well to be honest, I am not a huge fan of navigate-on-autopilot. I have a 2 hour commute per work day (1 each direction), most of it (~90%) on highway where I use autopilot extensively. But, NoA, I do not like and mostly due to its choices of lanes. It does not want to use the HOV lane, which is dumb when the traffic is heavy (usual) on the main lanes. It also wants to got out to the slowest lane ~3km before the exit, which is again dumb as it gets behind slow trucks. Then it starts overtaking them, sometimes very close to the exit, yet another dumb thing.

While AP itself is getting better, even lane changes requested by indicator is quite good now (used to be finicky, undecided, taking too long), but the lane choice logic of NoA is far too primitive. I am sure it is more useful for driving routes that you do not know much and would need to rely on the navigation to know which interchange to take. However, for routine daily driving I can make much better lane choices based on experience. For this reason I rarely use NoA, once in a while after each sw update to see how they improved it, but so far I am not impressed so I stick to basic AP on daily use.

Do you have Use HOV selected in the navigation menu?

Dan

 

[bdy0627]

Well to be honest, I am not a huge fan of navigate-on-autopilot. I have a 2 hour commute per work day (1 each direction), most of it (~90%) on highway where I use autopilot extensively. But, NoA, I do not like and mostly due to its choices of lanes. It does not want to use the HOV lane, which is dumb when the traffic is heavy (usual) on the main lanes. It also wants to got out to the slowest lane ~3km before the exit, which is again dumb as it gets behind slow trucks. Then it starts overtaking them, sometimes very close to the exit, yet another dumb thing.

While AP itself is getting better, even lane changes requested by indicator is quite good now (used to be finicky, undecided, taking too long), but the lane choice logic of NoA is far too primitive. I am sure it is more useful for driving routes that you do not know much and would need to rely on the navigation to know which interchange to take. However, for routine daily driving I can make much better lane choices based on experience. For this reason I rarely use NoA, once in a while after each sw update to see how they improved it, but so far I am not impressed so I stick to basic AP on daily use.

In reality, using AP instead of NoA for easy highway driving is probably the way to go for most drives currently. With current AP tech, all it really takes is an occasional flick of the turn signal when you want to change lanes. I would love less nagging since I always have my hands on the steering wheel but it does not detect it unless I hang the weight of my arm on the wheel. Honestly, the frequent nagging because it is not detecting my hands is what I would like to see improved most right now.

 

[TradingInvest]

I wish the car would stay a bit closer to the one in front on the most aggressive setting for dense highway driving. There is too much room for drivers to cut in front simply because AP allows more space than human drivers, so it is the pathway of least resistance.

What's your setting for the following distance?

 

[bdy0627]

What's your setting for the following distance?

I nearly always use 1 on the highway unless conditions are poor. I use 4 on rural 2 lane highways.

 

[TradingInvest]

In reality, using AP instead of NoA for easy highway driving is probably the way to go for most drives currently. With current AP tech, all it really takes is an occasional flick of the turn signal when you want to change lanes. I would love less nagging since I always have my hands on the steering wheel but it does not detect it unless I hang the weight of my arm on the wheel. Honestly, the frequent nagging because it is not detecting my hands is what I would like to see improved most right now.

I use NoA all the time, love it. I like it that I don't need to worry about missing the next exit, the car will plan for it. The car is too eager to switch lanes, so I adjust my speed to match the car in front of me, unless that car is too slow. According to Elon, the hand requirement maybe removed after a while.

 

[jerry33]

I wish the car would stay a bit closer to the one in front on the most aggressive setting for dense highway driving. There is too much room for drivers to cut in front simply because AP allows more space than human drivers, so it is the pathway of least resistance.

So do I [allow more room than average]. I'd much rather have the idiots in front of me where I can watch them, than behind me where they're likely to screw up and I'll be caught in the middle. The road is not a race track, no medal for arriving five seconds early.

 

[mekberg]

Anecdotally I've seen reports of ~50-mile commute legs on highways going 100% intervention free, and most of the interventions I've seen reported are actually voluntary, for cases where NoA is too timid in higher density traffic.

In principle HW3 might already be coast-to-coast capable... (!)

I drove from Madison to Chicago on NoA this past weekend. Most, if not all, of my disengagements were due to NoA being too timid in otherwise aggressive traffic.

 

[Mobius484]

Sooo, question for you experts.
If you had $45000 to spend, which option would you chose:

1) Buy a Model 3 (with FSD) and use it as part of the Tesla Network to generate money for the next 5 years.
or
2) Buy TSLA stock and sell in 5 years.

#2

I have enjoyed watching my AP 2 car learn over last 2 years. I find the process of developing the NN’s fascinating. EAP is plenty for me as I really don’t live in a city where I use it that often, car is too much fun to drive, so I spent my FSD upgrade money on stock. I’ll prob buy FSD some day when I buy my next Tesla....

 

[Todd Burch]

There is lot’s of talk about NoA performance on HW2. I think this will be irrelevant shortly.

The new chipset will allow vastly more neurons in the network, vastly larger matrices to solve, and give dramatically improved performance.

What’s neat is that none of us have seen HW3 perform yet.

I don’t expect the investor event to show a polished, fully automated experience. But I believe it will be able to show general handling of stoplights, intersections, and be able to navigate from one place to another (perhaps with some safety driver intervention).

Even if Tesla can do a demo with a few interventions for unusual cases while getting from one end of town to the other, this will show how far along Tesla is. Zero intervention should not be expected or needed at this point. It’s the pace of innovation that matters. Tesla has gotten where they are, from scratch, in just a few short years. That’s amazing.

Tesla already has the system in a deployable package. It’s already on cars being sold. Tesla is collecting or has the potential to collect more data than anyone else.

Tesla’s solution is vision-based and therefore cheaper than LIDAR systems.

If they can even demonstrate that they are roughly on a path to autonomy within a few years, I think the stock will see a big uptick.

Look at what the hugely-valued Waymo has going against it (and I don’t mean to pick on Waymo, their research has obviously pushed the boundaries of autonomous driving):

1. Their system is not packaged for production. Sensors hanging all over the roof. Not reasonably deployable in a production vehicle without significant repackaging and a deal with auto manufacturers.

2. Includes LIDAR, which raises the price of the system.

3. Relies on geofenced, detailed mapping data in a small geographic area which means it doesn’t scale and can only be used in a limited area.

NONE of these problems apply to Tesla. Their system is already packaged and deployed, they avoid the extra cost of LIDAR, and it’s a general solution.

If Tesla can even show that they’re getting close—even if they have to take over a few times during a demo—that will still be worth a lot to the share price, I think.

 

[VValleyEV]

I would love less nagging since I always have my hands on the steering wheel but it does not detect it unless I hang the weight of my arm on the wheel. Honestly, the frequent nagging because it is not detecting my hands is what I would like to see improved most right now.

I agree, but I don’t see a way out of this without some hardware upgrades, because all it can detect is torque. I experiment with my arm resting on my leg/knee and just a short length from there to the wheel, with a couple of fingers on the wheel and a little wrist torque, trying to keep it very small but constant without having to think about it. I think some folks just fold their arms and wait for the blinking nag, but that just seems wrong.

Aside: It did get better when my front half-shafts were replaced, easier to keep the wheel torqued on AP, along with tighter steering overall. This was done under warranty because I noticed some “shudder” on hard acceleration, and they immediately knew it was drive shafts. Pretty sure this was because mine was an inventory car with 8k miles on it, so lots of test drives with the hammer down. And I am told the shaft design has been updated with this in mind, so my new ones should be fine if I continue to drive “aggressively”. Hard to not, it is so much damn fun.

 

[KarenRei]

That's my experience seeing a dozen of so of these a day around my neighborhood. Makes me wonder if they are really focused on self driving, or more into generating data that they can sell to other AV companies

It sounds like they're doing the classic cheat to decrease their disengagement rate: manually driving the places where the cars are most likely to balk, and auto-driving where they'll be least likely to.

 

[mongo]

It sounds like they're doing the classic cheat to decrease their disengagement rate: manually driving the places where the cars are most likely to balk, and auto-driving where they'll be least likely to.

They could definitely be cherry picking routes.
However, my understanding is that the only disengagements that need reported are driver intervention for safety, and system self-disengagements. So balking (won't turn left) wouldn't count anyway ...

Here is my issue with the whole "appreciating asset" comment. It sounded to me that Elon was referring to the ability of the car to improve through OTAs. I don't think it was in reference to the Tesla Network. I think people are approaching this the wrong way. Our cars appreciate in their functionality, usability and usefulness. Whether they appreciate in monetary value is reliant on many other factors. Just my interpretation.

Dan

If they kick the price of FSD up 5k, cars that already have FSD increase in value 5k(ish).

 

[Todd Burch]

I hope the Apr 22 event will explain this question:

Say I buy a Model 3 and elect to "put it in the Network" so that it can make money as a ride-sharing vehicle.

Question: what happens if/when the FSD makes the 1-in-a-million mistake and does something that causes injury or property damage or worse? Who is responsible? Tesla? Me? The passengers? Say it is a really obvious case: the logs show the driverless car drove, at full speed, straight into an 18-wheeler crossing the road, the car simply didn't see the side of the trailer, and drove under it, passenger dies, etc.

Now scale it up to say 5-10 freak 1-in-a-million (heck, 1-in-a-billion) edge-cases a year, resulting in fatalities or serious injury. Is Tesla going to own this risk? Will individuals putting their car in the Network own the risk? Will Tesla indemnify such owners?

Inquiring minds, etc.

If there is a safety driver, the safety driver will be responsible.

The driverless scenario will simply not happen until statistics prove that such risks are miniscule.

Imagine you’re an insurance company. If Tesla proves that their automated system is less likely to have an accident than 90% of drivers, and less likely to have a fatal accident than 95% of drivers, would you take that robot on as a customer?

Of course you would, because you would lower the claims rate of the company as a whole.

AI can beat that statistic. Technically if AI proves to only have an accident rate on par with human drivers, they should be able to go on an insurance plan like any other driver.

Due to 100% concentration and vigilance, I have no doubt that AI systems will prove to be better.

 

[humbaba]

I agree, but I don’t see a way out of this without some hardware upgrades, because all it can detect is torque. I experiment with my arm resting on my leg/knee and just a short length from there to the wheel, with a couple of fingers on the wheel and a little wrist torque, trying to keep it very small but constant without having to think about it. I think some folks just fold their arms and wait for the blinking nag, but that just seems wrong.

Aside: It did get better when my front half-shafts were replaced, easier to keep the wheel torqued on AP, along with tighter steering overall. This was done under warranty because I noticed some “shudder” on hard acceleration, and they immediately knew it was drive shafts. Pretty sure this was because mine was an inventory car with 8k miles on it, so lots of test drives with the hammer down. And I am told the shaft design has been updated with this in mind, so my new ones should be fine if I continue to drive “aggressively”. Hard to not, it is so much damn fun.

One of the frustrations with NoA on my recent trip was lots of straight line driving causing the detector to be upset by the lack of resistance. Kinda hard to do when you are going straight -- actually, really, truly straight -- for extended periods.

But I also don't see a way around it -- even with hardware upgrades I'm not sure there's anyway to really reliably tell that the driver is attentive.

This is, in fact, a fair part of the reason I paid for FSD. As long as I can treat FSD like a nag-free AP I'll be happy.

 

[jerry33]

One of the frustrations with NoA on my recent trip was lots of straight line driving causing the detector to be upset by the lack of resistance. Kinda hard to do when you are going straight -- actually, really, truly straight -- for extended periods.

But I also don't see a way around it -- even with hardware upgrades I'm not sure there's anyway to really reliably tell that the driver is attentive.

The interior camera should be able to tell.

 

[MarcusMaximus]

One of the frustrations with NoA on my recent trip was lots of straight line driving causing the detector to be upset by the lack of resistance. Kinda hard to do when you are going straight -- actually, really, truly straight -- for extended periods.

But I also don't see a way around it -- even with hardware upgrades I'm not sure there's anyway to really reliably tell that the driver is attentive.

This is, in fact, a fair part of the reason I paid for FSD. As long as I can treat FSD like a nag-free AP I'll be happy.

FYI, my solution, since I figured out it works is: keep only 1 hand on the wheel, at ~9 o-clock. It just needs to detect an unbalanced torque on the wheel, so anything of weight on one side but not the other will keep it happy indefinitely.