Lidar lost Tesla the robotaxi?

[Misterbee]

I‘s a shame that the fleet of consumer owned Robotaxis looks like it will not come to fruition anytime soon. I had hoped to put my M3 to work for me.

 

[2101Guy]

the results of this poll are quite telling..

 

[t3sl4drvr]

Funny that people still believe this narrative when there are actual robotaxis driving on actual streets with actual passengers with no-one on the driver seat. And none of those actual robotaxis use a lidar or are made by Tesla.

Damn. Meant to write: And all of those actual robotaxis use a lidar and none are made by Tesla.

 

[DrChaos]

Vision has higher resolution than lidar. That's the problem with lidar and why Elon ditched it, the resolution is too low to be of any use. It's only good for telling you if "a large object" is at a certain distance from you.

No, it's not present because good enough long range lidar for autonomy (high frame rate and long distance) is expensive and ugly, and nobody would buy a personal car with it at present.

And Elon never 'ditched' lidar as there was never a Tesla sold with it. They ditched radar, probably because of a chip shortage but Elon lied about the reason (model S and european cars still got it). Autopilot without radar sucked for a year but now it's good again, at least in California.

 

[Dan D.]

View attachment 843903

Easy enough to test. @DirtyT3sla how about it?

String a white sheet across your rural driveway and draw a simple perspective road with marker. Doesn't have to be fancy just make it look roughly like a road when viewed from 50ft away. Don't even paint it, just a few heavy lines. See what the car does. Try drawing an intersection where none exists.

 

[damonbrodie]

The items in the poll are skewed to a specific perception.

(X) Tesla will win autonomous driving because they chose a superior approach with hardware and the latest AI tech.

 

[t3sl4drvr]

Tesla will win autonomous driving

This thread is about robotaxis, not autonomous driving in general.

Lets also try avoid vague "superior approach with hardware" and "latest AI tech". Those just do not mean anything:

1) Tesla AP HW3 is 5 years old and the sensor suite is the weakest in the industry. They might end up working for L4, but no-one knows at this point. It could also be that they are forced to upgrade to modern HW to catch up.

2) Everyone in "AI" uses the "latest tech" - new impressive deep learning research is published almost daily: every significant tech company follows and contributes to what is happening. Tesla is one of them.

 

[damonbrodie]

This thread is about robotaxis, not autonomous driving in general.

Lets also try avoid vague "superior approach with hardware" and "latest AI tech". Those just do not mean anything:

1) Tesla AP HW3 is 5 years old and the sensor suite is the weakest in the industry. They might end up working for L4, but no-one knows at this point. It could also be that they are forced to upgrade to modern HW to catch up.

2) Everyone in "AI" uses the "latest tech" - new impressive deep learning research is published almost daily: every significant tech company follows and contributes to what is happening. Tesla is one of them.

Robotaxi and autonomous driving are the same thing from the car from a driving systems perspective. Robotaxi layers on the "taxi" part in the software that deals with payment, hailing, etc.

At least one of the options on the poll should have been "Tesla updates to HW4 + the new Samsung HD cameras" - since we know that they signed a 7 figure deal with them.

This whole lidar/radar discussion is strange to me. It feels to me like a lot of it is fanboy/anti-fanboy stuff with Elon. Also some people maybe feel they are being ripped off because they got cars with radar and that got "taken away" from them. I dunno - but its really weird.

Tesla (Elon, and every Tesla employee that has publicly spoken on the matter) have all said that they are 100% in on Vision only because it is "better". I have no idea if "better" is because it actually is better or if it is simply a better fit for the skill set that Tesla currently employs, or some other reason. What is clear though is there is ZERO reason to believe they will make any decision to back away from Vision only.

 

[glide]

Robotaxi and autonomous driving are the same thing from the car from a driving systems perspective. Robotaxi layers on the "taxi" part in the software that deals with payment, hailing, etc.

At least one of the options on the poll should have been "Tesla updates to HW4 + the new Samsung HD cameras" - since we know that they signed a 7 figure deal with them.

This whole lidar/radar discussion is strange to me. It feels to me like a lot of it is fanboy/anti-fanboy stuff with Elon. Also some people maybe feel they are being ripped off because they got cars with radar and that got "taken away" from them. I dunno - but its really weird.

Tesla (Elon, and every Tesla employee that has publicly spoken on the matter) have all said that they are 100% in on Vision only because it is "better". I have no idea if "better" is because it actually is better or if it is simply a better fit for the skill set that Tesla currently employs, or some other reason. What is clear though is there is ZERO reason to believe they will make any decision to back away from Vision only.

Better = better for Tesla’s bottom line

Not better for the end-user.

 

[OxBrew]

Putting aside robo taxi/autonomous debate, vision only will always fail in fog/heavy rain/snow.

6-8 months of the year, there's a big chance of bad weather in the PNW, and similar places all over the world.

I just want enhanced safety in bad conditions, (radar or lidar head's up imaging of obstructions in the fog or rain) with or without any autonomous help.

This is such a huge blind spot, I'm kind of in shock it's not being worked on. (it is, just not by Tesla)

Solving for clear day well painted roads is maybe 50% of the solution, maybe less. The last x% (10?) of the solution will take 90% of the effort.

Bad weather won't be solved this decade, unless Tesla adopts something beyond vision only.

 

[t3sl4drvr]

Tesla (Elon, and every Tesla employee that has publicly spoken on the matter) have all said that they are 100% in on Vision only because it is "better". I have no idea if "better" is because it actually is better or if it is simply a better fit for the skill set that Tesla currently employs, or some other reason.

Better = The only cost viable way of getting "FSD capable" HW in cars 5 years ago

What is clear though is there is ZERO reason to believe they will make any decision to back away from Vision only.

If competitors start winning PR game after shipping L3 or L4.

 

[t3sl4drvr]

vision only will always fail in fog/heavy rain/snow

If not worse than a human, people will be ok with that: "I would not be able to drive in this snowstorm, no wonder my robotaxi did not arrive"

 

[johnnycnote]

Tesla's vision only is very impressive and seems to be improving rapidly. My money is on Elon making the right long term decisions with or without radar, lidar, or whatever else comes up

 

[hw13]

Damn. Meant to write: And all of those actual robotaxis use a lidar and none are made by Tesla.

They rely on HD maps.

 

[damonbrodie]

If not worse than a human, people will be ok with that: "I would not be able to drive in this snowstorm, no wonder my robotaxi did not arrive"

The sensors on my Hyundai go crazy when I'm backing up in the snow.

To be honest, I don't think any form of robotaxi will be running in a snowstorm. Certainly not for several years beyond any initial introduction.

 

[t3sl4drvr]

They rely on HD maps.

And that is bad how?

• Cars still operate on roads without HD maps
• Cars generate maps while they drive

Even Tesla (10.69) started to rely on maps to get cars to behave correctly in intersections.

 

[2101Guy]

look at that poll.....

 

[hw13]

And that is bad how?

It's not robo anything. It's essentially a railway car.

 

[Kablooie]

Based on the videos I've seen of the latest Tesla FSD Software, 10.69, I think Tesla may actually reach level 5 in a year or two. I haven't thought so up to now. I have FSD version 10.12 which is very flaky and far from being autonomous. But 10.69 implements a whole new 3D visualization technology and suddenly it looks like it may have a shot. It still needs fine tuning and more features such as understanding hand signals and ability to understand signs posting different rules for different times of day but It looks like it may have jumped way ahead of the others as of Sunday night.

 

[Supcom]

Based on the videos I've seen of the latest Tesla FSD Software, 10.69, I think Tesla may actually reach level 5 in a year or two. I haven't thought so up to now. I have FSD version 10.12 which is very flaky and far from being autonomous. But 10.69 implements a whole new 3D visualization technology and suddenly it looks like it may have a shot. It still needs fine tuning and more features such as understanding hand signals and ability to understand signs posting different rules for different times of day but It looks like it may have jumped way ahead of the others as of Sunday night.

From what I have seen, I would say that is very optimistic. There are still many things yet to be implemented and the pace of development does not seem to be sufficient. New FSD technologies may be a good thing, but still may not be sufficient to provide a robust solution. But, I haven't gotten 10.69 yet, so will remain skeptical regarding any schedule predictions.