Lidar vs Camera revisited

[Terminator857]

... we are going to get self-driving cars to be safe enough to deploy in the chaotic world.

Do you think that is true for level 3 on limited access freeway in stop and go traffic? Baby steps.

AI is a party trick and it gets us to a level we didn't easily attain before (in translation, for one example).

Life is a party trick invented by random chemical reactions with a tendency to self replicate.

big data was 'needed' but it still runs into walls and wisdom seems to indicate that its still yet another intermediate step in how we try to understand thought and mindful computation.

Yes a long ways away from FSD. Maybe next decade. But level 3 will be awesome and some crappy level 4 might be great.

the AI data model is sloppy, too expensive and its not going to get us there.

Moore's law can handle the expensive part. Every two years the cost halves. For machine learning it is improving faster than that because of improvements in software.

 

[eipi10]

My take is: who cares? Tesla hasn’t been able to program autopilot to handle two merging lanes in a sane, sensible matter. The fundamental problem is that the computer is stupid. That’s not going to change any time soon. It’s good at identifying fixed lanes and staying X feet behind the car in front of it, and that’s about all.

 

[marketisopen]

@Bladerskb you understand that tesla's whole approach of why they don't need pre generated HD maps is because their system generates said HD maps on the fly.

Why waste resources when the car can do it in realtime and better accuracy then pre-loaded HD maps. Nothing can beat realtime lol.

I just got FSD beta, and while the driving "Logic" is not there yet, they definitely have the vision system down. Its insane how good the vision system is at building the world it sees around it and how accurate it is.

@Bladerskb you understand that tesla's whole approach of why they don't need pre generated HD maps is because their system generates said HD maps on the fly.

Why waste resources when the car can do it in realtime and better accuracy then pre-loaded HD maps. Nothing can beat realtime lol.

I just got FSD beta, and while the driving "Logic" is not there yet, they definitely have the vision system down. Its insane how good the vision system is at building the world it sees around it and how accurate it is.

You answered your own question without a map the logic is more difficult.

 

[AKAlex]

I just got FSD beta, and while the driving "Logic" is not there yet, they definitely have the vision system down. Its insane how good the vision system is at building the world it sees around it and how accurate it is.

This appears to be the case to me as well. Beta currently seems to perceive the world very accurately. However it's logic on what to do with what it sees is not quite there yet. I don't see how LIDAR would fix this problem. To my eye, LIDAR in this context appears a bit like a red herring with a ton of institutional inertia behind it.

As far as maps, I wonder if data collected from cars is being used to update the vector maps? As in, is the vector world, as perceived by the car being fed back to the mothership if it does not match the current map data? This seems like a more effective way to solve map data issues than another gov agency...

 

[EVNow]

I just got FSD beta, and while the driving "Logic" is not there yet, they definitely have the vision system down. Its insane how good the vision system is at building the world it sees around it and how accurate it is.

Yes - perception isn’t the main issue anymore. As time progresses they will anyway get better at it. Now they need to
- make a lot of progress on the planner. I’d call the current system rudimentary - POC level …
- institute a way to correct 2D maps

No need to throw the baby out with bathwater like the concern trolls advocate.

 

[Bladerskb]

Nio's LiDAR supplier Innovusion plans initial annual capacity of 100,000 units

Innovusion's long-range LiDAR production line in Ningbo has been installed and is in quasi-mass production, according to the company's CEO.

cnevpost.com

 

[Terminator857]

Nio's LiDAR supplier Innovusion plans initial annual capacity of 100,000 units

Innovusion's long-range LiDAR production line in Ningbo has been installed and is in quasi-mass production, according to the company's CEO.

cnevpost.com

From the article: Even at 10% reflectivity can detect objects 250 meters away.
I wonder how much reflectivity flat black or matt black has. Seems to be a common car wrap.

 

[sjg98]

we have 'occasional crash' right now with human drivers but almost never, other than learners, do PB.

autonomous or even semi, should not be released until the PB problem can be managed, somehow. perhaps that's a learned stretch of road where the user says 'no! dont ever do that here'. right now, tesla does not learn from user driving, not like that. that would be at least one bandaid.

but really, if the feature is not ready, its not ready. and really - for most of the world, its not ready. I live in the bay area where its over-trained so I benefit from that. tbh, its one reason I bought mine; I knew I'd have a better driving experience than most other geo's simply due to tesla density here.

but I get the fact that PB happens a lot outside the bay area. that, to me, means, the tech is far from ready for widespread deployment.

I fear the day when teslas can be rented and cost similarly as ice rentals. regular people (lol) are just not ready for this. enthusiasts just BARELY are.

I normally describe Tesla Autopilot as, if anything, being unsafe in how conservative rather than aggressive it can be. That is - braking too hard/unexpectedly, weird hesitations at merges/ramps, etc that are going to cause it to get rear ended.

Similarly the following distance even set to minimum often leaves too much space and you get jacked up NY/NJ drivers using it to cut you off.

Lastly the hesitation in lane changes / luxuriously large space it expects people to leave it means I do not use AP in situations where I need to do a lot of highway interchanges/merges in a short number of miles as the car will never be able to get over on its own. I've had it half commit to a lane change and then swerve back into the lane when there is no apparent danger around me. Meanwhile I have also seen it happy to continue merging into a lane that has someone else merging into it from behind.. at which point I have to intervene. I end up using the lane change handling more as a nice party trick that allows me to monitor my mirrors and 360degree surroundings while the car does the actual change for me. This is not "auto" or "self driving".

The phantom braking really gets me because it happens in all sorts of unpredictable situations, meanwhile I have had to do numerous slam-the-brakes manual interventions getting cut off at highway speed where Autopilot did absolutely nothing. Lately I've found very aggressive PB on HOV lanes in flat lighting or even after dark, on wide open lane with no one to sides or in front, nor any overpass casting a shadow. I drive on LIE (I-495) in NY here where the HOV has a particular merge lane pattern for a good 30 miles of highway, and the car will be OK with 90% of them and then do hard PBs on 10%.

AP has improved a lot in the almost 4 years I've had the car, but the rate of improvement has certainly slowed.. and was never enough to give me confidence in actual FSD inside the next decade. I'm actually tipping more in the direction of AP being more likely to cause me an accident than to prevent me from having one. I mostly see it as a convenience feature in long (20+ mile) stretches of highway driving.

 

[okcazure]

From the article: Even at 10% reflectivity can detect objects 250 meters away.
I wonder how much reflectivity flat black or matt black has. Seems to be a common car wrap.

That should be ideal. That's the ideal surface for FLIR and I suspect it would be similar for Lidar. I'm not an expert, just my observation from a similar technology