Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
Yes, it's' more or less this. Many poisons can affect the brain, and in particular cause its tracking between (say) balance and the visual system to get out of sync. The best survival response is to get rid of the poison, hence vomiting. So it's not the ear per se, but the brains interpretation of different sensory inputs. That's why you get sea sick ..your ear is telling the brain one thing and your eyes another. Same for too much alcohol. Kids are more susceptible as they are less tolerant (another survival thing).

So when you look down in a car, the car body and (e.g.) book is telling you that you are stationary, but your ear says you are swinging around, and yeah you get sick. The best mitigation is to look out the window AHEAD of you, so you "see" the same thing your ears are telling you. Same thing for sea sickness, get on deck and look toward the bow of the boat at the horizon.
Cool, I almost added before that this issue of the brain is probably more likely in folks (like me) already neuro-divergent (ASD Level 1). That seems to be the consensus that I've encountered... lucky us.

Oh, and if you're in a bus, especially toward the rear, don't try to look forward, find a hill out your window...
 
Last edited:
The "preferences"(like Chill mode etc) typically become inputs to the NN, with appropriate training data for each mode, though this can be expensive if there are many such settings (one reason Tesla keep them simple).

Dont forget the NN doesnt just get camera data .. it also get map info, accelerometer, GPS, compass etc.
While conceptually you are correct that different data could be used to train different networks in the different modes, doing this with just two toggle settings means you have to train and test 4 separate networks. Then multiply by the regions that need their own networks and this quickly becomes an unscalable problem.

No way is Tesla doing this. To maintain the chill/average/assertive modes + minimal lane change toggle and 3 regions (NA/Europe/China) means you now have at least 18 separate networks to train. Then those all have to be tested. And they all need vast amounts of data. The compute time and extra storage required alone would be prohibitive.

I think the settings were maintained to support the continued v11 highway stack. For the next few years, I think the only settings we’ll see are ones that can be applied to the output of the neural network by traditional C++ guardrail code.
 
While conceptually you are correct that different data could be used to train different networks in the different modes, doing this with just two toggle settings means you have to train and test 4 separate networks. Then multiply by the regions that need their own networks and this quickly becomes an unscalable problem.

No way is Tesla doing this. To maintain the chill/average/assertive modes + minimal lane change toggle and 3 regions (NA/Europe/China) means you now have at least 18 separate networks to train. Then those all have to be tested. And they all need vast amounts of data. The compute time and extra storage required alone would be prohibitive.

I think the settings were maintained to support the continued v11 highway stack. For the next few years, I think the only settings we’ll see are ones that can be applied to the output of the neural network by traditional C++ guardrail code.
I don’t think they would necessarily do this either. Maybe for regions but not for every toggle. There should be other ways to control it. Something like how you would prompt a LLM: « please keep the Gs down and try very hard not to change lanes, ok? ». There’d still be an increased test burden but maybe not so much with the training. So much I would like to know about this architecture…

I think they already do some supervisor-style guardrails or overlays with map data because it will not try to do right turns on red here most of the time. Still doesn’t understand arrows, though.
 
Last edited:
The important question to me is, does it EVER hard brake unnecessarily with someone dangerously close behind? At least since V12? Like, has anyone here had that happen?

Not sure if you are talking about phantom braking? For me that is mostly a thing of the past.

Otherwise it will brake hard when needed regardless of closely following traffic. There's been recent videos of following traffic needing to divert to the roadside to avoid rear end contact. It will also brake hard for yellow lights even if traffic is following close. It will also cut in front of quickly closing traffic from behind.
 
Not sure if you are talking about phantom braking? For me that is mostly a thing of the past.
Did you catch my post from a few days ago? Hard braked from 45 to about 3 on the deserted road heading to my house. No other cars.
Thank God it isn't doing it to me on the highway, and this was just one time, but PB is what it was.
 
While conceptually you are correct that different data could be used to train different networks in the different modes, doing this with just two toggle settings means you have to train and test 4 separate networks. Then multiply by the regions that need their own networks and this quickly becomes an unscalable problem.

No way is Tesla doing this. To maintain the chill/average/assertive modes + minimal lane change toggle and 3 regions (NA/Europe/China) means you now have at least 18 separate networks to train. Then those all have to be tested. And they all need vast amounts of data. The compute time and extra storage required alone would be prohibitive.

I think the settings were maintained to support the continued v11 highway stack. For the next few years, I think the only settings we’ll see are ones that can be applied to the output of the neural network by traditional C++ guardrail code.
Not true, though I understand your point. You are not training/testing 4 (or N) distinct networks. You are training ONE network to respond correctly to numerous inputs, the "mode" being just one of very many more.. The "mode" is no more special than (say) the distance from a car in front of you .. they are just training data inputs. It's important to understand that from a training standpoint there is no difference between a mode input and a sensor input .. its all just data that is fed to the NN and trained to generate the correct output .. the network doesnt "know" the mode bits are special. Two mode "bits" doesnt increase the network complexity by 4x because NNs dont scale by 2^N (where N is number of inputs) .. otherwise we'd never get any useful work out of them at all.
 
Yes, focusing out of the vehicle at some distant stationary object works, and as soon as possible after you realize you're not feeling so hot. I spent my childhood not doing well in cars and my parents just gave me dramamine. Funny thing is when I first tracked a car, I didn't understand why I was getting sick. The first few times actually. I'm a slow learner. Since I'd already solved the problem for while scuba diving (Scopalomine patches), I started wearing the patches to the track. Helped a lot. But ultimately I don't like the wear-and-tear on my car from tracking. My motorcycle instructor has been trying to talk me into doing a non-sportsbike track day for years, but I'm chicken. Of getting sick, not of the riding part.

Do you guys know why some people get motion sickness? Maybe the physicians here can correct me if the story is apocryphal, but I believe it: the vestibular system in the inner ear is really sensitive to chemical imbalances, so we've evolved such that if we were to ingest a poison (bad berries or mushrooms or whatever), the inner ear reacts first and thus such a reaction triggers a reflex to "purge the poison". The body is being fooled into thinking it's been poisoned.
Um. No. At least, I haven’t heard that one.

Been at sea. Got seasick once, early on, then got over it; not unusual, this happens to a lot of people.

The issue, as I understand it: one is in a ship/vehicle, ship is moving up and down, pitching fore and aft, rolling side to side, and it’s all very irregular. Also, put oneself far to port or aft and thus add in the sensation of going up and down in a speeding elevator as well.

Finally, put oneself in some interior compartment or corridor, with one moving or not, but no view of the outside. One’s eyes says that the walls, ceiling, and floor are absolutely rigid with 90-degree angles everywhere and nothing is moving, absolutely contradicting one’s inner ears and kinesthesia that say one is being tossed around like a pea in a random pod. A lot of people get nauseous in this situation where the eyeballs and everything else disagree.

Eventually, most get over it. It’s an odd sensation: in one’s mind’s eyes one sees the walls tilting and pitching back and side to side; and one is able to stand, walk, run, and even sit at a table without falling out of one’s chair. Some people, actually, never do figure out the mental trick. They’re the ones for whom sea travel isn’t recommended. I understand that drugs like Dramamine help; I suspect that class of drugs are psychoactive.

If on a ship, going to a weather deck where one can see the horizon helps. But I’ve seen newbies to the whole situation where, horizon or no, being tossed around like a chip for the first time when one is used to one’s whole world being stable results in tossed cookies. It’s kind of a learned trick.

Once one has learned the trick, one can still get nauseous under the right circumstances. But, no kidding, one’s sense of balance tends to really improve. My son tells us that, when he was little, he really liked having me around on a subway car ride when we had to stand. He found it safer and more secure to hang onto me rather than any old metal pole-I was more stable than them 😁.
 
Not true, though I understand your point. You are not training/testing 4 (or N) distinct networks. You are training ONE network to respond correctly to numerous inputs, the "mode" being just one of very many more.. The "mode" is no more special than (say) the distance from a car in front of you .. they are just training data inputs. It's important to understand that from a training standpoint there is no difference between a mode input and a sensor input .. its all just data that is fed to the NN and trained to generate the correct output .. the network doesnt "know" the mode bits are special. Two mode "bits" doesnt increase the network complexity by 4x because NNs dont scale by 2^N (where N is number of inputs) .. otherwise we'd never get any useful work out of them at all.
Hi, Drtimhill --

> You are training ONE network to respond correctly to numerous inputs, the "mode" being just one of very many more..

I don't understand how this would work in an any sort of "pure" ML context. It seems to me if you want "Chill", you'd have to identify a bunch of "chill" drivers and train on that.

Mmm ... I take it back, I guess you could set up the "reward function" (NB, I have no idea what I'm talking about) to measure "Chillness" but that seems ... hard, and maybe a bad idea? Like, for "Smooth", you could punish hard braking somehow, but are you really sure you want to do that?

This is a confession of ignorance and not an argument.

Yours,
RP
 
Um. No. At least, I haven’t heard that one.

Been at sea. Got seasick once, early on, then got over it; not unusual, this happens to a lot of people.

The issue, as I understand it: one is in a ship/vehicle, ship is moving up and down, pitching fore and aft, rolling side to side, and it’s all very irregular. Also, put oneself far to port or aft and thus add in the sensation of going up and down in a speeding elevator as well.

Finally, put oneself in some interior compartment or corridor, with one moving or not, but no view of the outside. One’s eyes says that the walls, ceiling, and floor are absolutely rigid with 90-degree angles everywhere and nothing is moving, absolutely contradicting one’s inner ears and kinesthesia that say one is being tossed around like a pea in a random pod. A lot of people get nauseous in this situation where the eyeballs and everything else disagree.

Eventually, most get over it. It’s an odd sensation: in one’s mind’s eyes one sees the walls tilting and pitching back and side to side; and one is able to stand, walk, run, and even sit at a table without falling out of one’s chair. Some people, actually, never do figure out the mental trick. They’re the ones for whom sea travel isn’t recommended. I understand that drugs like Dramamine help; I suspect that class of drugs are psychoactive.

If on a ship, going to a weather deck where one can see the horizon helps. But I’ve seen newbies to the whole situation where, horizon or no, being tossed around like a chip for the first time when one is used to one’s whole world being stable results in tossed cookies. It’s kind of a learned trick.

Once one has learned the trick, one can still get nauseous under the right circumstances. But, no kidding, one’s sense of balance tends to really improve. My son tells us that, when he was little, he really liked having me around on a subway car ride when we had to stand. He found it safer and more secure to hang onto me rather than any old metal pole-I was more stable than them 😁.
In one respect I agree it's possible for some of us who are motion-sick-prone to train ourselves out of it to some degree. I did it while riding buses as a member of the L.A. City Youth Band. We were bussed to parades all over SoCal, and I really wanted to be able to read a book in the bus. I learned how to do it, but it was with my brain, not just by "getting used to it". I had to keep part of my awareness on the motion of the bus at all times. Fortunately it's not hard to read and maintain awareness of a vehicle's motion. As soon as there was any kind of bump, I just immediately looked out the window at something distant until the oscillations dampened enough, then went back to my book.
 
Well of course it does, but when it’s travelling so far to the right, it’s often an endless ping ponging weaving left then right, then left then right. I’m not really sure WHY it wants to try and hug the right side of a marked lane so much. As I’ve said, I’ve seen this pretty much from the getgo ~ 2 years ago? Or is it 1.5? I can’t explain it but what concerns me MOST is why does THIS car think it needs to do this, or that this is correct? Does anyone else see this much offset?
After the last update, I notice my car is riding far to the right side of the lane all the time. Very annoying. Also, if I set the speed +5 mph, many times the car will not respond unless I physically press the accelerator. Don't believe there is anything to do but wait for the next update.
 
After the last update, I notice my car is riding far to the right side of the lane all the time. Very annoying. Also, if I set the speed +5 mph, many times the car will not respond unless I physically press the accelerator. Don't believe there is anything to do but wait for the next update.
Mine also biases too far to the right. I especially don't like it during right turns but also when there’s parked cars. Zero chance if someone opens their door or pops out.

I don’t mind the accelerator overrides so much but it’s annoying that I can’t move the car so much as an inch with the steering wheel without disengaging.
 
Great discussion on motion sickness.

I never had it in cars but on a cruise ship, I learned my headache and drowsiness was actually mild motion sickness. Seabands and if extremely rough, dramamine, work for me. (Writing that makes me wonder if my Seabands could allow me to work in the back of the car, I should try that this summer.)

I don't know how much tendency to it is hereditary but my father and grandfather were sea captains so mentally I tell myself I'm made of the 'right stuff'. And in general never have an issue.

Interestingly, dramamine knocks me out for at least 24 hours on land if I take it when I have a tummy upset, but on a ship, it just works fine and I continue on, just slightly slower than usual. I figure it is because the drug is actually counteracting something. Ginger works just as well for me so I've switched to that. (Again, I should try this in the car - I really miss being able to work in the back seat.)

I was taught to look out at the horizon on a boat, so I figure the back seat of the car being so limited in the front view is the main problem with me feeling sicker in the back than the front.

Finally, re sick kids in the car, we were taking a grandkid home and suddenly I heard the sound of car sickness. We were on the highway, but light enough traffic that I was able to climb from the front passenger to the back seat, then reach into the far back to grab the towel we keep there for wiping the back camera off after a rainy drive, before buckling up and starting to clean up. I don't travel light so within reach of me or the driver were baby wipes, water, cloths, plastic bag, and once the car was stopped in a parking lot, a clean sheet to wrap around the wet child for warmth and to cut the smell. Car seats are awful to throw up in, you can't move at all, so it just puddles over you, poor toddler.

In the plus column for the MY, it had enough room that I could climb from front to back (I'm not small, nor flexible) but it was also small enough that I could reach into the rear well in the back to grab the towel. Which means, at least in that case, it is the perfect size car.
 
Re: lack of rear-ending accidents reported here or in the general media.

I was one of those who predicted the media would catch wind of a lot of fender benders (at least) caused by FSDS and it being used by people unused to supervising unreliable software.

I've been pleasantly surprised that there hasn't been such reporting. There's a lot of wheel curbing and summon scrapes reported here but not a lot involving other vehicles while driving on roads.

Both phantom braking and other strange behaviours seem to happen when it doesn't put the car at risk. Either that or the general driving population is excellent at giving teslas a wide berth so they avoid hitting the teslas doing weird stuff in front of them. Obviously, it has to be the former because I doubt anyone thinks the general driving population is excellent at anything.

So V12 is obviously ready for wide roll-out even if it is far from perfect.
 
A couple of notes from the morning drive today.

1) FSD does not GAF about obstructed visibility. The roads around here are in a grid with lots of parked cars. There are also a lot of a-holes that gun it across the road without checking for oncoming traffic. When I honk at them, they'll either act oblivious, or honk back at me, depending on the driver. Happened to me today where a driver gave me the finger, even though he's the one that pulled right out in front of me forcing me to brake and go around him, and I didn't have a stop. I'm not even laying on the horn -- just a quick boop to let them know I'm there.

In a situation like this:

a) If there is a large truck or something else obstructing your view of cross traffic, then even if you have the right of way, you need to slow down and be prepared to brake hard. FSD proceeds on its merry way without a care in the world.
b) If a car is stopped and there is a gap in front of it, you need to slow down and be prepared to stop. Odds are, there may be a car about to turn left across your path (right from your POV), or a pedestrian may be crossing. Either way, this is a bad time to be going full speed ahead.

Of course, other times it will brake cautiously in places where you don't need to because visibility is good, so there at least I can do an accelerator override.

2) I'm getting a lot of false FCWs on right turns in this build. My FCW setting is set to "late", but FCW will still highlight a parked car in red. It's completely false since I would never even come close to hitting that car, even if I did nothing else at the time of the FCW.
 
Eventually maybe, but it's not ready for that

Maybe not ready to perform perfectly, but certainly more ready than an unconscious driver.

We know that Tesla has been working on networks that watch the driver-facing camera for signs of drowsy driving. If they can get a network that can identify an unconscious driver with a reasonable degree of accuracy, it would probably save a few lives.