Underwater robots must compute locally because communications are slow

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 27:49

   like it's just kind of default state salt water and the ocean is very hostile environment what's harder than people realize when you're building a robot that lives in the ocean uh you know corrosion is like and pressure are like huge like external forces but you know corrosion you can get around it by coating uh you're doing like anodizing um the the surface of the metal um which gives it like some resistance pressure again you can mitigate i mean you know going very deep sea which would be 6000 meters is what we call full ocean depth i mean the ocean goes to about 10 000 meters in its deepest parts but full ocean depth in a technical term is the 6000 that's very difficult we operate currently in like the 1,500, up to 1,500 meters.

2. 28:29

   I mean, we've been crush-tested to 2,200 meters, but we're typically operating in that 1,500-meter range, which is where most of the important economic activity today happens. In the future, I think much more will happen in the D.C., predicting in critical minerals. But today, a lot of the economic activity happens there. So pressure is one problem. I think the biggest problem, though, is comms underwater.

3. 28:51

   I was going to ask that. Yeah, yeah. So how do you communicate with the maker? A lot of it has to run locally. You have to do a lot of sense, because you can acoustically... Because communications are very low bandwidth. Yes, you can do like 10 kilobytes per second over like five kilometers. Over like HF or what? Acoustically, ping, like a dial-up modem. And then just like being far away from, being far away from anything, you know, like is very difficult, like, you know, being far from someone that can repair you or connect and recharge you.

4. 29:25

   So that's why the Surface vehicle is such an important part of our stock because you pick it up, you put it on deck, it wirelessly charges, Leviathan will wirelessly charges our rewarder vehicles and then connects us to the internet so that we can get a live diagnostic on the vehicle, what's working, what's not working and then if someone needs to come in and do an intervention. I mean, in the future, the Surface craft will actually be able to repair the vehicle and have replacement modules and kind of vending machine style.

5. 29:49

   The Surface vehicle is autonomous too. That's what's right. Yeah, it's pretty cool. And it's connected then to Starlink, so we can see. And Starlink is a huge enabling technology for us. So that's one part of it. That's why we go for the Surface vehicle. Another thing is, yeah, we have around 100 times more onboard compute than any other autonomous underwater vehicle. We run a full NVIDIA 5090 onboard with an Intel CPU.

6. 30:10

   And that's been very important to our autonomy stack. Because, again, if you want to do anything intelligent robotically underwater, you need to be able to sense make to a large degree. Okay, but basically, does the Mako, if it's doing some mapping or whatever, do you send it down with a fully built plan and program, and it's basically, it is running the full loop of sensing and deciding locally on the GPUs on the device?

7. 30:36

   Yeah, like right now, the autonomy is relatively rudimentary. We haven't gotten into very complex autonomy behaviors just yet, but not many of our tasks require it just yet, but yes. But to the extent that it is autonomous, it is fully locally autonomous. yeah exactly it's like uh it will be like draw a polygon that's like map this area okay yeah auto generates a lawnmower track and then it'll have some like rudimentary okay if i see a rock let me just like go above that or if i see like something else go around it yeah and then