Anthony Levandowski on lessons learned at TC Sessions: Robotics+AI



so you're speaking of hardware though you are known for your work on lidar and yet pronto is not going to be using lidar and I'm wondering why is it is it because overs are you just staying away from it because of the way Moe uber a lawsuit or is there some other reason why lidar is not in the picture with pronto yes really good so that's a good question too so I would say so I don't have any restrictions I'm not doing lidar but I do have a restriction personally of not doing things that I know aren't gonna work so in the past you know I understood I could see what we were doing and basically back in 2000 you say 9 or 10 you could see that the lidar x' didn't have the performance for what you needed because you couldn't see far enough to actually safely have the system reacted of software i was i've seen safely how the system's see what's going on to have a quick reaction it turns out that even with lidar what's missing today is not seeing more accurately or seeing further what's missing is understanding what is happening to those vehicles around you and being able to predict what the motion is so you can react accordingly and be safe and not just predict what their motion is gonna be but what they think you're gonna be doing and so we we mainly focus on that because that's the that's the building block that is unlocking the in the future now I'll say I was a huge proponent of lidar and well that was not my quote but I definitely believed that that was actually the unlocking factor you know and wiser person than me and I'll use some humble pie here said that lidar is a crutch so you'd agree with Elon Musk he said and he's been ridiculed for that a bit he's right for calling lidar a crutch and that's what you believe as well I believe that as well and I don't believe that because I can't do it because I'm happy to do that it's just not where I think the actual value and you could look at that right so if you were to like analyze all the disengages that people have done and try to break it down like what's the actual real reason behind it is the real reason because the brakes weren't redundant or the steering wasn't redundant no in fact that's like zero point zero percent of all this engages our because there was no failed operational system all of them were software failure and they're mainly a software failure you know on mature companies at trying to understand what the vehicles gonna doing or the pedestrians around you and you having a misunderstanding or miscommunication with them and so that's the that's where the values gonna be now we could have this conversation in a couple of years and be like what that wasn't the problem the problem is this other new thing that we don't know about but you know like all scientific process you make a hypothesis you test it and you're wrong and then you move on you know like before Galileo we used to think that the earth was the center of this you know Sun went around us and then he realized it was opposite that was a major discovery back then and they're like oh we're gonna this whole like Inquisition and all kinds of thing that could relate to you a little bit about that and and so you know I would say going forward you know I'm all about finding where the problem is and solving that and right now the problem is not better lasers the problem is perception not prediction your prediction it's interesting because you talk about the willingness to discard things that don't work and yet you you really did work on a single premise for a decade or so and I'm wondering if this is gonna sound crazy but was the lawsuit and everything that happened sort of a blessing did it cause you to rethink things or were you are ready starting to rethink that maybe this level for first approach isn't the way to go yeah so think that it's hard once you're in the middle of something and you're like dedicated fully to it it's very difficult to get perspective and see like you think you're there but actually you're missing this other path that's gonna block you completely so I definitely would say that's taking a step back and you know taking some time to reflect on my life and you know things I could do better and you know we start to church yeah well that's not when the church was started but that's uh that was while I was at Google in 2015 but to think about really what is the actual problem and and and be able to start from a clean slate it's very difficult to do at a company specifically when you're dealing with the day-to-day about like this team and that team they'll get along and all kinds of other problems at that point it's very difficult to like really set the strategy what you need to do you just focus on executing along that path so well know if it was a blessing disguise or not but you know I think then I'm pretty hopeful about the future and I'm optimistic about what we're doing so while you have now pursuing pronto there's dozens of companies that are still pursuing this level for and I'm just wondering in your view now and I wonder how they would respond to this but they're not on stage so we can't ask them but do you no longer believe that level four is achievable safely in the short term well so I would say as I encourage people to have different approaches because I could totally be wrong and so it's great that other people have different points of views and you know what I would say is like the true metric for when a v's actually get to be deployed is safety right so safety is the only metric that determines when a navy will come out not whether they use lidar or not or car maker or software vendor and so I don't personally think that straight to level 4 is gonna be something that is doable because you'll need to have a very long time before you actually solve the problems and by providing technology that has evolved and gets better over time you'll be able to like chip away and all the problems people use an example of if you go into the moon you know building a ladder doesn't really help you well the thing is that we're not going to the moon we're just trying to drive to the corner store and back and so we shouldn't be building you know that that's an analogy that while you should be building the right technology you're at the right strategy it's not necessarily that you can't make an evolutionary step to get there and I think that the evolutionary step will uncover a lot of lessons learned that will make it you know safer and better so what should we do to appease our AI overlords prior to the singularity apocalypse phenomenal question well you can sign up for you to ask any course and learn how to do that I'm sure there's other classes that you can do as well no I mean ultimately I think it's a funny question but you know progress is inevitable and you know you want to make sure that you helped create the progress and understand where that's going as opposed to deny it or try to cage it up or you know saying it's more dangerous than nuclear weapons you know and just be fearful basically AI has a huge opportunity to connect us in a way that you couldn't before and we can build that in a fearful way or in a way that has bias or we can build that in a way that's gonna be helpful for us so don't be scared is that you know three-letter answer or three word answer

14 thoughts on “Anthony Levandowski on lessons learned at TC Sessions: Robotics+AI”

  1. Elon is right. Lidar will work in a simulated environment. Its just too dumb, they will need Camera/AI tech

  2. so after this guy is legally banned from working w/ lidar he has the revelation that its not necessary….

    https://www.androidheadlines.com/2017/05/ubers-levandowski-legally-banned-working-lidar.html

  3. The amazing thing is not just Elon was right but he saw exactly what it is even before he started to execute. Other very smart people still need to fool around with it for a long time before learning the truth.

  4. “I think if somebody started today and they were really good, they might have something like what we have today in 3 years, but in 2 years, ours will be 3 times better." – Elon Musk

  5. While of course lidar is not reliable enough because it can't detect metal, shapes, texture, or luminosity. But clearly Elon Musk has it wrong as well, using video, because his Teslas have hit many trucks already, using video.

Leave a Reply

Your email address will not be published. Required fields are marked *