LIDAR - Safest car on the road potential?

[gaskiller]

These "opinios" are based on the evidence of every car that exists in china.

We have 20 brands with self driving in china. 20 of them. They're all bad. Out of 20 of them, only 2 of them are "decent" which puts them at the level of FSD v12. Almost all these cars have more compute and sensors than tesla does.

Rivian claims they will do unsupervised robotaxi in first half 2028. That's the same time xpeng says they will do unsupervised robotaxi. Xpeng is actually a leader in china in autonomous driving and rivian is years behind. If Xpeng gives a date more than 1 year out, it's because they really don't know when they will have a product good enough for unsupervised.

Tesla had 15 years to make this work with just cameras……still doesn’t.

 

[Macgreiner]

I watched Aptera's CEO talk about why they went with Commai and not LIDAR for driver assist. Apart from costs he mentions the power draw from an active LIDAR is not insubstantial.

 

[Aka-none]

I watched Aptera's CEO talk about why they went with Commai and not LIDAR for driver assist. Apart from costs he mentions the power draw from an active LIDAR is not insubstantial.

So safety does have a price

 

[ke2]

So safety does have a price

The fact that LIDAR has a not insubstantial power draw is actually useful information and could have as much to do with the decision as the cost of the LIDAR (we don't know how factors were weighted)

 

[Adnillien]

The fact that LIDAR has a not insubstantial power draw is actually useful information and could have as much to do with the decision as the cost of the LIDAR (we don't know how factors were weighted)

The LiDAR sensor on the Air uses about 15W of power. Not really significant compared to even HVAC let alone the power train.

 

[Bdr100]

Tesla had 15 years to make this work with just cameras……still doesn’t.

But FSD currently works better than any other commercially available ADAS system on the market (outside of Waymo with it’s $100K+ sensor suite). Right now, if you want a car capable of point-to-point ADAS here in the US, there is only one choice -Tesla, so to say that it is not working is not accurate. Yes, I really hope that Lucid/NVIDIA are able to leverage their Lidar to create a system better than FSD but only time will tell. Based on Lucid promises of AR HUD and actual delivery, I am a little hesitant to take Lucid at their word when promising future tech. Most of the cars in China that everyone are talking about have multiple Lidar around the car. Lucid, Rivian, Volvo, Mercedes are planning to use just one small forward facing Lidar.

 

[Aka-none]

But FSD currently works better than any other commercially available ADAS system on the market (outside of Waymo with it’s $100K+ sensor suite). Right now, if you want a car capable of point-to-point ADAS here in the US, there is only one choice -Tesla, so to say that it is not working is not accurate. Yes, I really hope that Lucid/NVIDIA are able to leverage their Lidar to create a system better than FSD but only time will tell. Based on Lucid promises of AR HUD and actual delivery, I am a little hesitant to take Lucid at their word when promising future tech. Most of the cars in China that everyone are talking about have multiple Lidar around the car. Lucid, Rivian, Volvo, Mercedes are planning to use just one small forward facing Lidar.

Ok so if Lucid were to follow in China’s footsteps they would own the entire production from people to raw materials.

Personally I’ve never purchased anything based on what some future version would be.

So far I believe, could be wrong, that fsd has killed or maimed more people than any other brand of car with any form of self driving. Yes they have more cars on the road and that impacts the numbers.

QQ for those in the know
If fsd is active and you decide to go somewhere else, and fsd remains on, how long before it disengages?

 

[Bdr100]

Ok so if Lucid were to follow in China’s footsteps they would own the entire production from people to raw materials.

Personally I’ve never purchased anything based on what some future version would be.

So far I believe, could be wrong, that fsd has killed or maimed more people than any other brand of car with any form of self driving. Yes they have more cars on the road and that impacts the numbers.

QQ for those in the know
If fsd is active and you decide to go somewhere else, and fsd remains on, how long before it disengages?

This has been said before, FSD has exponentially more miles on it’s system than any other in market, so yes, it will have more incidents. Also, most of those incidents were with drivers asleep at the wheel. If I were to engage my Gravity’s handsfree DDP2 while on highway and then fall asleep, I guarantee you it would kill me.

 

[Nunya]

If fsd is active and you decide to go somewhere else, and fsd remains on, how long before it disengages?

You can say (or type) "go to Subway on Main St" while driving and it responds in seconds regardless of current destination. You can also say "add a stop at Subway" and it will make that an intermediary stop along the route.

Of course you also just touch the brake or turn the steering wheel to disable FSD instantly and drive anywhere. Pull into Subway and resume your regular FSD drive.

Not sure that answers your question though. If "FSD remains on" you either need to put in a new destination, waypoint, or disengage.

 

[Adnillien]

This has been said before, FSD has exponentially more miles on it’s system than any other in market, so yes, it will have more incidents. Also, most of those incidents were with drivers asleep at the wheel. If I were to engage my Gravity’s handsfree DDP2 while on highway and then fall asleep, I guarantee you it would kill me.

More likely Gravity would execute an in lane stop because driving monitoring recognized that you were asleep.

 

[Bdr100]

More likely Gravity would execute an in lane stop because driving monitoring recognized that you were asleep.

My point was that IF I was able to fool the system into not recognizing that I was asleep, like most of the Tesla FSD accidents. I would be willing to bet that if Lucid owners were to be given the choice of keeping the current DDP2 system “as it stands right now, not based on future hopes and promises” or switch to Tesla FSD, most of us would take the FSD.

 

[martinjlm]

GAWD I tried to stay out of this one. And maybe after this post many of you will tell me I should stuck with that.

I think that part of the issue is Lidar is being painted with a broad brush. As if all sodas were cola and there was no ginger ale, 7-Up or Dr. Pepper.

If lidar is so key to good ADAS, then why has Volvo abandoned it (for now). I have always wondered just how much Lucid actually uses the current lidar on their cars. I have a feeling it was more of a “lidar is getting cheaper so lets put one on our cars and hope we figure out how to optimize it eventually”. Putting lidar on cars also helps justify the very high DDP and DDP2 upgrade costs even though DDP currently is about a good as similar systems from competition that cost 1/3 what Lucid charges for DDP(2).

The key will be when (if) the NVIDIA Alpamayo goes live in Lucid. Hopefully we will have better data in another year or so when more consumer cars with lidar based point-to-point systems start racking up miles on the roads.

The issue with Lidar implementation is two-fold. Maybe more, but definitely two headwinds are cost and integration. The integration piece is where the flavors of soda come to play. All Lidar approaches are not created equal and at least up until now, no automaker has cracked the code on the correct approach to implementation that can survive a business case challenge.

Well if ev’s were so great then gm, ford, and stellantus would not have reduced their commitment to them

So we may as well just go backwards.

The point is that companies do all kinds of things for reasons not always apparent to outsiders.

Three different statements in one sentence. If commitment to EVs was salsa, Stellantis would be mild, Ford would be medium and GM would be spicy. GM CEO Mary Barra has stated on multiple occasions, in public, full diaphragm that EV is the future. The issue is timing to get there. GM has backed off on the 2035 timing originally stated because the current administration knee-capped the incentives for building EVs and batteries in North America. Without those strategic measures, making a short-term wholesale swap from cash register 5-digit profit generating large SUVs to expensive to build EVs would be foolhardy. So now GM is in the mode of carefully navigating continuing the profit machine while still launching new EV product. Very good new EV product.

Ford still has Bill Ford pushing a green agenda but his voice is not as loud as Mary Barra's. Ford is content with being a fast follower. If Ford had truly abandoned EVs we would never see Fathom. Some might say "Fathom was too far along in development to cancel". I would counter by pointing out that the refresh of Lighning was further along, as was Blue Oval City manufacturing complex. Ford pulled the plug on both of those but kept Fathom going. They understand that the market is more appreciative of a low cost small pickup than a full size EV pickup with the compromises that have been surfaced by Lightning, Silverado EV, and Sierra EV.

It's marketing. They put it in and convince you the lidar is the reason they'll have self driving where others will fail. Consumers eat this up. They believe rivian has some advanced self driving hardware when the memory bandwidth (the real limiting factor) is just as bad as everything else on the market.

There is also a technical answer that in the future if some generic AI gets ridiculously efficient and you can run on a very weak computer, then more sensors could possibly unlock self driving in a car that could theoretically not be sensor limited

in the case of say tesla if there is a camera blind spot or camera weakness that can never be overcome

To me this logic is nonsense as you can always manufacture a car faster than you can unlock self driving for consumers.

If the bold part were true then we would be seeing one of the wildest intentional squandering of R&D funds and integration engineering spending in automotive history and auto executive heads would roll. Execs would be fired, companies would fold. Instead, development continues and integration efforts each learn from the most previous failed approach. This is the way technology development sometimes works in the auto industry. I could give countless examples of similar situations on the combustion side, many of which I've had direct hands on. Spend some time researching HCCI and then ask me why we don't see it in market now.

Tesla had 15 years to make this work with just cameras……still doesn’t.

Ahhh, but it does. I've come full circle from despising early deployments of FSD because of a couple near catastrophes to actually admiring it now that it has improved to very high 90 percent ranges and drives better than most humans. And they've done that with cameras only. Super counter-intuitive from my experiences in product development roles, but results are really hard to argue with.

I still fault Tesla, though, for not immediately disallowing engagement of FSD if any cameras are blocked and/or at least notifying the driver that system performance is degraded when cameras are blocked. They do that now, but they did NOT do that as recently as a year ago when my bike on bike rack behind my Model Y confused the camera system into believing it was being tailgating. This resulted in random speeding up and dramatic lane changes to try to avoid the tailgater. Now if I try to engage FSD with a bike on the rack I get a notice that the camera is compromised and FSD is not allowed.

But FSD currently works better than any other commercially available ADAS system on the market (outside of Waymo with its $100K+ sensor suite). Right now, if you want a car capable of point-to-point ADAS here in the US, there is only one choice -Tesla, so to say that it is not working is not accurate. Yes, I really hope that Lucid/NVIDIA are able to leverage their Lidar to create a system better than FSD but only time will tell. Based on Lucid promises of AR HUD and actual delivery, I am a little hesitant to take Lucid at their word when promising future tech. Most of the cars in China that everyone are talking about have multiple Lidar around the car. Lucid, Rivian, Volvo, Mercedes are planning to use just one small forward facing Lidar.

As Shakira once said "results don't lie". That is what she said, isn't it?

It very well could be (and should be) that with optimized sensor fusion and integration, camera + radar + lidar may deliver an even safer and more effective self-driving experience. Logic leads to that conclusion. But that optimization has not happened yet. At least not in a consumer affordable manner, and at the same time, Tesla's FSD has raised the performance bar. Such an integrated system has to come out of the gate with 99.9999% effectiveness. FSD still makes mistakes, but in my family's experience over the last year or so, those mistakes are few and non-critical. Turning into the wrong driveway at a destination..turning right on red where there are signs posted clearly stating it isn't allowed. Stuff like that.

Compared to Waymo? Not taking that cheese. It's like comparing an Apple Watch to a Rolex. Same function, different reason for being.

 

[Aka-none]

This has been said before, FSD has exponentially more miles on it’s system than any other in market, so yes, it will have more incidents. Also, most of those incidents were with drivers asleep at the wheel. If I were to engage my Gravity’s handsfree DDP2 while on highway and then fall asleep, I guarantee you it would kill me.

I wasn’t aware that Lucid encouraged disengaged drivers yet that is exactly what is implied by Tesla and fanbois. Right on teslas front page are videos of people in the drivers seat dancing or completely unable to take over in an emergency. Like grandma is all prepped for an emergency.

Unless I’m mistaken about the gravity it’ll stop the car if you fall asleep or fail to pay attention. Does fsd?

 

[martinjlm]

...

QQ for those in the know
If fsd is active and you decide to go somewhere else, and fsd remains on, how long before it disengages?

In my recent experiences (recent being past 12 months or less) I'm more likely to disengage FSD than it is to disengage itself. I honestly can't remember the last time it disengaged on its own. Typically when I disengage it, it's because when we get home I'd rather back the car into the garage than pull it in nose first. There's not a way that I or my wife know to change that tendency. That and there are times in parking lots when I want it to park in a place different from what it chooses.

If your question is about a change in route it's kind of a red herring. You simply tap "End Route" and it will continue in Self-Driving mode with no route planned. If you want you can voice activate a new route and it grabs that immediately.

 

[Aka-none]

In my recent experiences (recent being past 12 months or less) I'm more likely to disengage FSD than it is to disengage itself. I honestly can't remember the last time it disengaged on its own. Typically when I disengage it, it's because when we get home I'd rather back the car into the garage than pull it in nose first. There's not a way that I or my wife know to change that tendency. That and there are times in parking lots when I want it to park in a place different from what it chooses.

If your question is about a change in route it's kind of a red herring. You simply tap "End Route" and it will continue in Self-Driving mode with no route planned. If you want you can voice activate a new route and it grabs that immediately.

Close but what I was interested in knowing is if self driving is active (anyone’s system) and I grab the wheel and steer the car elsewhere does self driving stay engaged or not?

I’m driving and I need to take over right now. What does the system do, how does it behave? And now I’m out of just the fsd response but any self driving car

 

[martinjlm]

Close but what I was interested in knowing is if self driving is active (anyone’s system) and I grab the wheel and steer the car elsewhere does self driving stay engaged or not?

I’m driving and I need to take over right now. What does the system do, how does it behave? And now I’m out of just the fsd response but any self driving car

Drops out immediately. That is one of the three methods for disengaging FSD.

1. Tap the brake or
2. Take over steering or
3. Tap the screen where it say it is in Self-Driving mode

 

[Need4Sleep]

Personally I like to think of the aviation industry. The reason why it's so safe is because of redundancies on top of redundancies. And it is within engineering's duty of care to ensure the safety of the passengers wherever they can.

I'll just leave this here:

Though the topic is about self-driving cars, there's a lot of points with respect to sensor-fusion w/ lidar vs vision-only.

Finally, the kind of data we need to see the incidence and crash rates between the two systems is very likely to be internal data that we would not have access to, and I think it's important to keep that in mind

 

[TahoeMichael]

But FSD currently works better than any other commercially available ADAS system on the market (outside of Waymo with it’s $100K+ sensor suite). Right now, if you want a car capable of point-to-point ADAS here in the US, there is only one choice -Tesla, so to say that it is not working is not accurate. Yes, I really hope that Lucid/NVIDIA are able to leverage their Lidar to create a system better than FSD but only time will tell. Based on Lucid promises of AR HUD and actual delivery, I am a little hesitant to take Lucid at their word when promising future tech. Most of the cars in China that everyone are talking about have multiple Lidar around the car. Lucid, Rivian, Volvo, Mercedes are planning to use just one small forward facing Lidar.

The $100K+ sensor cost figure for Waymo is way way way out of date. That's from the early days, when all of the sensors were low volume products, and one step away from engineering prototypes. Waymo doesn't disclose a lot of specifics about their driving platform, but they have said that the cost of their sensor suite has dropped dramatically from those early days. This shouldn't surprise anyone familiar with the cost curve of hardware products. LIDAR technology has improved dramatically in the last decade, and the volume of sensors manufactured has also dramatically increased. We're at the point where low end LIDAR is so cheap that there are a bunch of cell phones with LIDAR sensors included.

 

[TahoeMichael]

I think a big reason why people disagree so passionately about whether FSD has failed or not is whether you're going off the name and view it as an autonomous driving system, or whether you go off how it's sold and view it as an ADAS. As an ADAS, it's amazingly capable, best available, although not perfect and I do still have some issues with it (as do a number of others here). As an autonomous driving system, it's a bit too much YOLO'ing for my taste, and pretty hard to argue that Waymo isn't substantially ahead of it. Obviously we have a new poster here who vehemently disagrees, but I think it's clear from Musk's hard charging nature that if Tesla could be operating autonomous driving the way Waymo does, they would be.