Ethan Goodhart on Developing Autonomous Vehicles at Stanford

into Sanford. Some of these lunatic parents look at the tables >> law. Okay. >> Of which states >> because they want to have representation from every state. Which state has the least number of >> Wow. >> And they they'll literally move to Nebraska. Ruin their lives for two years, three years. >> Georgia's got to be doing pretty good. I feel like Georgia's competitive in ter like West Virginia. That's where you'd go. Idaho. Those are >> even It's true though, right? Stanford states. >> I' Yeah. I mean that's not the reason why we're in Atlanta. No, but uh >> but you have heard this talk about >> Yeah. >> Yeah, I have. >> Yeah. This this is the lunacy. Okay. So, uh what do you computer science, electrical engineering, double what are you doing? >> Yeah, pure computer science. Uh this is my first like real hardware project. >> Got it. So, uh self-driving uh you know law, my position has been 20 different companies of note are going to get there in the same 24month period. putting Whimo out of it just starting this year and next year into 2028 my belief Ethan was 20 people figure it out. I think I'm good like in the same 24-month period we'll see 20 people figure it out. And when I saw Ethan's project it said to me I might have underestimated it by a magnitude. It might be 2,000 people figure it out in the same 36 months just to extrapolate a little bit. tell us what you built and how long it took to build and just the fidelity of it versus the brittleleness of it. >> Yeah, for sure. Uh and it probably the best project I've ever worked on my life. Uh first hardware project. We got the first prototype up in like two, three weeks. So we had a golf cart that could drive with some like very simple models. Um, obviously not as advanced as a a whim or a Tesla, but uh, yeah, we've kind of been working on it for a couple months after that. And then >> this is a standard golf cart. And you built some kind of rig to go on or a harness, but not a harness. Yeah. >> It's a physical harness that's going over the uh, steering column. Yeah. >> Yeah. Exactly. It's it's so it's retrofit. Um, we didn't modify any of the like internal electronics. >> This is all displayed on top of the golf cart. >> Yeah. >> Um, so it's like three front cameras, >> three back cameras, and then >> no lighter. All cameras like Elon. >> No LAR. >> Okay. >> Yeah. >> Um, and so you take a stock one, you put cameras on it. What is the hardware stack? Because, you know, Elon's been working on this for a long time. uh and building essentially the compute locally on the car to do exactly what you're doing here. So yeah, I see there's kind of a box on the floorboard. So I'm assuming you have an Nvidia Spark or something down there doing this. >> Yeah, we we have a Nvidia Jets and Thor that's wired directly to the battery of the golf cart. >> Got it. Um so that uses a decent amount of power. >> Yeah. Um, and the golf carts are notorious for going what, 50 miles in a day, 30 miles before charge. So, how do you deal with that? Do you buy a bunch of anchor batteries and just bolt them in? >> So, right now it the the range is a little bit limited. So, it's definely something we have to figure out uh in the future, but it it's still very much in like testing phases. >> On the top, you got a Starlink. I see. Um, >> Starlink Viny. That's right. >> Yeah. Fantastic. Uh, what is that doing? Uh so that allows us to do two things. Uh it allows us to test out cloud inference. So we're exploring running some delays and world models in the cloud >> and then it also allows us to have good connectivity to write code and ship code live to the golf cart while we're driving around campus. Got it. >> Um so like I often work on it outside of my dorm. Spend a lot of all nighters in spring quarter working on this. And I can like ship code live while I'm on the golf cart instead of having to like go back into a dorm room or classroom somewhere. >> Okay. Aut I mean listen it is autonomous is on golf courses. Uh people don't know this but a lot of the uh grass being cut right now law and even in uh snow country a lot of the snow being moved is being done by autonomous robots. >> The team will go um find a video or two about this. But even here in the great state of Texas, Eth and I drive around and I see in the planned communities robots cutting grass. Um, and that's in like Hill Country. That's here today. They're doing a lot of this on golf courses. Uh, you made this as a proof of concept, but you know, you're right next to Sand Hill Road. There's so much money slloshing around. Is the idea to like go to golf courses and say, "Hey, we literally can help you bring the golf courses back. So here's the snow one. Check this out, Nathan. This thing's incredible. >> That's awesome. >> Um, and so, you know, snowb blowing is one year when we had a lot of storms up in uh Tahoe at my ski house, I got an incremental $12,000 bill because they're paying the people who shovel snow $70 an hour. They're literally getting paid $150,000 a year, Lon. And we had so much snow that it was going to collapse the houses. Therefore, there was no choice but to outbid everybody to save our houses. Anyway, these things are becoming uh standard operating procedure. Here's one on a golf course um using uh Toro, TOEFL Pro, Anonymous, and these things are really easy to set up. People love them. What's your plan? I mean, are people asking you to build this for a golf course is because you might have a situation where somebody stops and leaves their golf cours do you see this as a business? >> I think it's uncertain what the future of of this was. This was just a fun side project, which I think it's absolutely incredible that something with the scope now can just kind of be a side project. Uh the future is uncertain, but there's definitely been a lot of interest from people in the golf world, also people just in the transportation world. So like people like Disney and >> other companies that have a lot of golf carts for transportation. Um I think there's a lot of use cases for golf cars and Yeah. >> What can you uh if you were to iterate on this twice and get, you know, obviously this is like a custom sensor pack, but let's say you had 10 million in funding, you know, you get to, you know, >> oh yeah, >> number 500. What do you think it would cost to retrofit these these golf carts cost 10 to 20k depending on how fancy dancy they are? What what do you think you can get the retrofit to to have autonomous golf carts in the real world? >> Yeah. Uh even the current version is quite cheap. So I won't say the exact number but feature >> Yeah. where you could see a retrofit version >> of a kit like this where you could sell for >> under a couple thousand dollar. >> Amazing. >> Wow. >> Incredible. >> What software did you use? There's been a lot of talk about Nvidia and Uber and um this open- source stack that they're providing because they want a Uber obviously strategically wants open-source driving. They want a fragmented market. They want every single car to have it so that Whimo or Uber or you know pick your company Neuro, Pony AI, Wave, AV Zuks don't run away with it. They want it to be fragmented. And Nvidia wants to sell compute hardware. So, it's good for them if the software is abstracted. Are you using their stack for software or did you write your own or are there other open source ones? Tell us about the self-driving software, the brains. >> Yeah, the software is by far the hardest part so far. Um, we've ended up using our own software. >> And so, we found that a lot of the open source models surprisingly were either way too large or didn't generalize well to the environment that we're driving on. um because it's a little bit different than the environment that most cars are driving on given like the pedestrian density and temperatures is very different. >> Um so there's also there's also the the question of the models being too big. So most of the models that have been published fine video so far are actually like quite large and hard to run on even the like $4,000 GPU >> and so we ended up using uh our own sort of approach and like custom approach. >> Love it. Listen, continued success. Uh if you make it into a company, just save uh save a slice for your Jcal. Uh, you know, I I'm I'm good for like, you know, if if I get on the cap table, I'm always good for like a ticket to I don't know, an all-in summit, something. I mean, it might be good for a retweet. So, this is what I got to do, Lon. I got to set I'm like a I'm like 55 year old man >> talking with 20 year olds now and trying to get a slice on the caption. What point is that desperate? >> Is it 65? >> No. >> Where it feels desperate or >> you just graduate to becoming more and more senior, you know, like where you're >> Okay. like venode. >> People want you on the cap table even more because of your incredible reputation and Yeah. >> Okay. All right. All right. So, I get to I'm in my Obi-Wan era right now, Ethan. That's a Star Wars reference. >> And then I guess Benode is in the Yoda