Monday, April 22, 2019

Elon might be right.

I know tons of people are bagging on Elon about his comments on LIDAR, but 99.9% of those people are completely unaware of how enormously large these data sets would be.

I have only recently become aware of this myself. I mean, you know it in the back of your mind, but these are real problems now that people have to compensate for. And even I'm not sure of the full impact. I mean, this data is crazy large.

So, I'm not sure if he's right or wrong. But I do know that most people talking about this really don't know what they are talking about. I am not even sure how they are going to store all this data. And I've been on the LIDAR bandwagon pretty early on.

I do love to shit on Elon. But I also like to be fair.


  1. He said -
    ‘Anyone relying on lidar is doomed,’ Elon Musk says

    “Lidar is a fool’s errand,” Elon Musk said. “Anyone relying on lidar is doomed. Doomed! [They are] expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices. Like, one appendix is bad, well now you have a whole bunch of them, it’s ridiculous, you’ll see.”"

    It goes on to say...

    "Elon Musk previously explained that he views lidar as a crutch for self-driving vehicles. For Tesla, cameras are the keys to the future and its CEO sees a future when cameras will enable Tesla to see through the most adverse weather situations.

    Andrej Karparthy, Senior Director of AI, took the stage and explained that the world is built for visual recognition. Lidar systems, he said, have a hard time deciphering between a plastic bag and a rubber tire. Large scale neural network training and visual recognition are necessary for Level 4 and Level 5 autonomy, he said."

    It's one thing to Lidar map a building or a small region, but Lidar is now going to have to experience some growing pains mapping large moving regions or cities which engineers are just now starting to face. These data sets will be massive. And they will add 5-30 grand in price for a car.

    I'm not really sure what the right answer is because they are just now starting to run into problems. I am watching like everyone else to see how this plays out.

  2. Yeah. Me either - but Elon is still right. Lidar doesn't see these things. In sand, fog, rain there is nothing for the laser to hit and return a signal. I have even heard of instances where Lidar could completely not see a dusty black car. Snow confuses the sensor. There are a lot of things Lidar can not see. It's just the facts. The lidar pulse has to make a return trip, so anything that can not achieve that process can not be seen.

    Be a hater for the right reasons. And be fair when the situation warrants it.

  3. Awww. See. Now that's nice. You don't ~have~ to. I'm not telling you what to do. But there are just a million other absurd things about him to pick on. In this instance he's not wrong.

    I was sort of sad to find out that lidar couldn't map the ocean. Ran into a surfer that CES had brought in, and thought it would be cool to have a surfer run the ocean waves with lidar. But apparently that is not a thing. And honestly a lot of Lidar systems use cameras. I'm going to tell you a little lidar secret. Lidar is overlayed with greyscale photography. then programmers have to infer the colors later.

    Now I've just ruined the illusion.

  4. Capital of Texas RefugeeWednesday, April 24, 2019 1:28:00 AM

    You still have an illusion or two, and now I'm going to ruin those as well.

    Look at the problems from systems theoretical points of view instead of from the view of end-user products and components, and you'll see a kind of history that repeats that isn't immediately obvious to most people.

    The reason for that is that they've grown accustomed to the systems around them, but they haven't stopped to think about how their "current state of things" actually came into existence.

    The post office, railroads, the highway system, the Eisenhower Five Star Interstate System (a.k.a. Neue Deutsche-Amerikanisch Autobahn, WW2 tech without the Nazis), air traffic control, swipe credit cards with 1950s magnetic tape tech, the Internet ... what's common about all of these big systems?

    All of these systems are relatively straightforward in terms of interfaces that people can deal with, and any complexity that they hide tends to be covered over with engineering that makes dealing with the problems a matter of repeatability, procedure, and maintenance.

    Any new systems must first overlay the old ones before replacing them, and they have to do this repeatably through procedures and maintenance that make them at least as reliable as what they're replacing.

    TCAS and ADS-B at first overlaid LORAN and then replaced it.

    E-mail is currently overlaying postal mail and has replaced most of it, leaving package services as the major remnant.

    Printed catalogs are still being made, but most of that activity has shifted to the Internet.

    In terms of form following function (at least as far as old American architectural myths go), you modify the function and you get a different form.

    Except that doesn't actually happen with anything the size of a system.

    Even in 2019, we still have recording technology from the 1950s on the backs of our credit cards. We still have "blinkers" on our cars instead of a variant of TCAS for vehicles. We have people making hand signals mandated by law on motorcycles that are moving fast enough to be a complete blur. LORAN may yet come back because it's deemed essential for national security. U-Line still sends me catalogs because the Internet isn't as fast as page flipping and picking up a phone if you have a lot of repeat orders.

    People tend to fuck up their long-term needs by hoping for short-term solutions that only accidentally scale up to long-term solutions.

    One main problem is that technological innovation does not scale quickly or directly to the size of a full system replacement. E-mail is enough of a problem that we still have the US Postal Service shuffling around envelopes. EMV chips (1980s tech) wear out chip card readers on what appears to merchants to be a surprisingly frequent basis.

    And so the unbelievable arrogance of all of these technology propagandists is that all it's going to take is a relatively simple fix to integrate new systems into old systems: add a LIDAR rig here and there, add a small fuckload of cameras, and shove Marvin the Paranoid Android's CPU stack next to the gas tank, it'll all be good.

    You want to believe good things about these people because they're so fucking sincere, but I don't believe any of it. I thought for a while that they must have some Magic Tech Smoke I wasn't seeing in Florida, but I found out that in many ways they are behind what we are doing in another line of business.

    And that's why I keep coming back to systems reinventions for completely different systems -- those are actually doable and should solve most of the end-user demands ...

  5. Capital of Texas RefugeeWednesday, April 24, 2019 1:37:00 AM

    US railroad systems are long due for a major systems refactoring, and one reason that's not being done is because the individual companies that own them are unable to cooperate to the extent required to implement new uses of inter-modal transport. Nothing's forcing them to change, and so all of the changes they've adopted (such as container transponders, smart signaling, and so on) have been piecemeal additions over an incredibly long time frame.

    What's needed is a relatively seamless way to link railroad systems and highway/interstate systems for inter-modal travel so people can be picked up in their cars and moved on railroads.

    Then you'll get the "sleeper car" you want because you would be able to sleep in your vehicle while you're pushing across California at 95 mph, a speed that most railroad systems in the US can either already handle or be upgraded to handle.

    At that point you won't care whether your car is "self-driving" because all of the long-distance irritating driving would be taken care of by the inter-modal systems.

    But instead of doing something involving the gradual overlaying of existing systems so you can eventually get to the results you want, Elon Musk wants to believe in Magic Cameras, and you wanted to believe in Magic LIDAR in order to imagine the same results.

    Putting all of the tech into the vehicles because the road networks and inter-modal networks have been deemed unfit or unable to change essentially burdens all of those vehicles with the problem of needing to be absolutely perfect from day one ... which, of course, has never happened pretty much ever for anything at any time in history.

    Extrapolating the complexity of the required systems changes yields the right answers, and most of the rest is essentially technological hope for people who hold close various technological religions.

    My main concern with all of this techno-hope is that it's going to blow out budgets for anything that actually has a better chance of working, but isn't as amazing or wonderful because that involves a bunch of relatively boring additions to systems that already work.

    And that's why you probably won't get your "sleeper car" either way you want that to be a thing: I'm actually expecting a fairly major tech implosion caused by all of the techno-hope chickens coming home to roost.

    Any further techno-hope illusions I should know about? :-)

  6. Well, I think you are inferring a lot of things about me that do not exist. "You want to believe good things about these people because they're so fucking sincere, but I don't believe any of it." That is your own bias. I'm not even sure how you got to this. I don't believe anything these people say. And I thought this blog was pretty forward about that. I believe what I SEE.

    Having said that - humans are accident prone. Inattentive. Not bright in a lot of cases. Other humans are complete assholes and walk right across the street without looking. Look down for a second and you might just hit one. So having some slight computer interference is acceptable. Because I"m too pretty to go to jail.

    I also believe that it doesn't matter what you or I feel about this technology. The horse is out of the barn, and they just aren't going to put it back. I've always said there would be a suite of sensors. I'm not bound by lidar or any of the others. I'm just stating facts of the real capabilities of lidar and it does not see everything, and it's not cheap.

    I do however like that if a car stops quickly in front of me before ai can react my Audi puts on the breaks so I don't die. Everyone accepts this as okay technology and the other stuff will eventually be the same. That what I see.

  7. LOL. It's not really a nerve - I just know people working in this field right now. You don't even have to take my word for it. Just google it. Or duckduckgo it.

    People are just not going to accept a car going up thousands of dollars just because they have lidar. You can believe me or not. You do you.

    Especially when cheap micro camera are now coming on scene enmass. There is going to be a lot of consolidation in the lidar sector. Mark my words.

    Love the song though.