Stefan (00:03):

Hello and welcome to Automate It, our weekly podcast, where we talk about automating all sorts of things, big and small. I, of course, am Stefan Seltz-Axmacher, the man with the name that everyone can't pronounce, and with me is, I don't know. Who?

Ilia (00:17):

Depending on the week, I'm either the shadow government or the puppet of the shadow government, depending how it's-

Stefan (00:22):

Someone's been on Twitter too much.

Ilia (00:24):

No. I've just been watching Inside Job, a great Netflix series.

Stefan (00:31):

With me and a big fan of Inside Job is my co-founder and CTO at Polymath Robotics, Ilia Baranov. And this is our weekly podcast where we talk about building robots. We start off with my favorite section of any podcast that you've ever listened to, which is where we draw two different cards and then come up with a robot on the spot for it. And then after that we're going to have a fun little change of pace from what we normally do.

(00:55):

What we've done so far, as we've talked about robots we've built in the past. We thought actually this week we turn it around and poke some playful fun at smart people we've met recently with a partially fictionalized account of how they've been trying to build some robots. So with that being said, Ilia, let's play the game that I stay up at night hoping to play where we pretend to start another robotics company. This time it's one where you are really eager on the setting and the business model that we're working on and what setting and business model is that?

Ilia (01:25):

We picked transportation this week.

Stefan (01:27):

Transportation. And I will draw a tech card and this is almost like the... I read it on TechCrunch, so we have to use it. How about this one here? Transportation and robotic arms.

Ilia (01:37):

Awesome.

Stefan (01:38):

So in this case...

Ilia (01:39):

Catapult.

Stefan (01:40):

... What we're going to talk through is what could we build that might be useful in the transportation industry with robotic arms with an emphasis on, of course, not just talking about the Toyota factories.

Ilia (01:52):

Yeah. So definitely first thing that comes to mind, you get in a seat. It's a large robotic arm. You tell it where you want to go, it gives you a parachute.

Stefan (02:01):

Okay. Naturally.

Ilia (02:02):

And then it catapults you in the direction you want to go.

Stefan (02:05):

That's one. And I did that. I'm interested in, so I like stairs. I like stairs. Escalators, as Mitch Hedberg once rightly pointed out, are just stairs with a little bit of jazz. And broken escalators are just called stairs. Elevators, that's so yester-year. What if I need to go from the ground or the plane that I'm currently standing on to a variation of different planes within 40 feet horizontally and 40 feet vertically?

(02:35):

I think a giant robotic arm that picks me up and places me in one of those places is the technology that, I mean, maybe I stand on. I think this could be useful. Say for example, you were building an airplane, you're in the Boeing factory, that big mile long building, and you want to go from screwing in a bolt to the bottom of the plane to screwing in the exact opposite bolt on the other side. You should stand on a robotic arm that moves you to the other side of the plane.

Ilia (03:05):

Maybe you're wearing your harness and the robotic arm is attached to the harness. So basically, you're just floating in space and the arm can position you arbitrarily.

Stefan (03:05):

I could be-

Ilia (03:05):

At any point.

Stefan (03:17):

... Like a boom camera. I could be-

Ilia (03:18):

Except a person.

Stefan (03:18):

Yeah, a boom camera, but a person and if we're going to do this, probably the ROI of course is how much I can move around, because ladders are just difficult.

Ilia (03:28):

Just slow.

Stefan (03:29):

Yeah. So move me around like a boom person. The controls problem on this I think is interesting, because I am not a pillow, in terms of what you tell. I'll say that. So moving me from one part in the plane to another...

Ilia (03:42):

Do you have to do it fast? Because that's always the trade-off. If you could do it slowly, you can get away with smaller motors and just a bigger gearbox.

Stefan (03:50):

Well, what would-

Ilia (03:50):

How fast do you need to go?

Stefan (03:52):

I mean, I'm thinking I'm screwing in a bolt on the bottom of the plane and I need to get out the other side of this plane. I don't know. What's a supermax? How is, maybe 40 feet in diameter? So I need to get on the other side of a 40 foot tube.

Ilia (04:05):

Well, I'm thinking if you compare it to stuff that exists. If you look at those buckets for electrical service. Electrician's cherry picker, I think.

Stefan (04:15):

Yeah.

Ilia (04:15):

Stands in the bucket gets picked up. It takes 20 seconds to go from 30 seconds to ground level.

Stefan (04:21):

And I get that transportation is not a factor, so I'm more probably trying to get somewhere. Maybe this is how I get into my tree house, or we all have tree houses because I spent a year in Southeast Asia, I saw this water village in Brunei, and maybe actually you use these arms as a way of getting from a boat to the house on stilts. That's actually the way to do it. Think of how much more efficient water taxis would be if you didn't have to pull up to a dock and you could just...

Ilia (04:52):

It just yanks you up there?

Stefan (04:54):

Yeah, maybe there's a large robotic arm attached to a water taxi.

Ilia (04:59):

Is it attached to water taxi or is attached to you? Making kind of like a Dr. Octopus?

Stefan (05:04):

Look, that-

Ilia (05:05):

We know that never ends up well.

Stefan (05:06):

Well, that's unreasonable. That's just not a thing that we should have.

Ilia (05:10):

I mean, power source is the question.

Stefan (05:12):

Yeah.

Ilia (05:12):

I mean how do you-

Stefan (05:14):

Listen, we just get one of those little round things from Iron Man. It's all gravy.

Ilia (05:17):

Yeah.

Stefan (05:17):

You know what I'm talking about? No, no, no, I think you have a water taxi. You have a robotic arm on it that has a platform. You stand on the platform and it moves, it puts you into the water.

Ilia (05:27):

But why? Sorry, why did we drift away from Boeing factory to pedestrian?

Stefan (05:31):

Yeah. To pedestrian. To close to manufacturing. Yeah.

Ilia (05:34):

Yeah.

Stefan (05:35):

So these water taxis, they tend not to be new boats. So it's probably a retrofit.

Ilia (05:39):

Water taxis. But why not? Why not? Okay. Why not use this to make the world more accessible for people who can't use stairs? Right?

Stefan (05:46):

When you think about it, a giant robotic arm that moves you from place to place is kind of like an elevator on the go.

Ilia (05:53):

Yeah. Yeah. It's a more flexible elevator.

Stefan (05:56):

Yeah. So see, you could have a bunch of these, almost like those genie lifts that are autonomous, driving through cities in places where they haven't made them very accessible.

Ilia (06:10):

That's right.

Stefan (06:11):

So really outside of the US, seemingly...

Ilia (06:14):

Been in the US since.

Stefan (06:15):

So you pull out an app on your phone.

Ilia (06:18):

Uber.

Stefan (06:18):

You say like, "Hey Uber, I have some stairs I need to get over."

Ilia (06:21):

Yep.

Stefan (06:22):

The self-driving genie lift drives to you through traffic and pedestrians. We'll see that's really easy.

Ilia (06:28):

That's the easy one.

Stefan (06:28):

Yeah. Yeah.

Ilia (06:29):

Problem. Yep.

Stefan (06:31):

Then it comes and positions a ramp free to roll onto, picks you up the 20 or so feet that the stairway is and then gets you to the other side. I mean actually, a cool thing is I went to College in Philadelphia, the rocky steps by the art factory. The art museum. You could have someone who can't climb stairs go up on top of the art museum steps by this robotic arm.

Ilia (06:56):

That's right. And the benefit of not being permanently mounted somewhere was you could have just one servicing a bigger area.

Stefan (07:03):

Yep. You have them go from baseline. Now you would have to wait for it to show up.

Ilia (07:06):

Yeah. Could you get it down to Uber timelines? Like five, eight minutes?

Stefan (07:10):

I think it depends. I mean if you think about it, what's this arm going to cost? What's a genie lift cost? My guess would be a genie lift is like $200,000. So maybe this arm is $600,000 to $800,000.

Ilia (07:24):

In that range.

Stefan (07:25):

Yeah.

Ilia (07:25):

Are you including the piece that moves it around autonomously?

Stefan (07:28):

Yeah, I'm considering the actual wheels that get it around.

Ilia (07:32):

Yeah. Yeah. So, you're right. You're talking about like 600,000, probably.

Stefan (07:36):

So, civil engineering projects are stupidly expensive. Saying this from San Francisco where we recently bought a 2 million dollar bathroom.

Ilia (07:43):

I was going to bring that up too.

Stefan (07:44):

So you could have, rather than a 3 million bathroom, you could have three different autonomous lifts to make spaces accessible that weren't previously accessible.

Ilia (07:53):

Even better. You make the arm just one payload. The other payload of this moving vehicle could be a bathroom.

(08:02):

And the problems swap out different payloads.

Stefan (08:04):

Yeah, that's true. You have a movable bathroom. You could have a... What happens if someone who's accessibility challenged only has one bathroom and they can't leave their house? You need to get bathrooms to them?

Ilia (08:16):

All sorts of payloads. I heard SF is considering allowing lethal force for robots. You could have a militarized payload on board. No problem.

Stefan (08:25):

Yep. Kind of anything that you want in a large movable arm in an urban set.

Ilia (08:30):

That's right.

Stefan (08:31):

So it's a really easy problem to solve. All we need to do is wait for general autonomy to build.

Ilia (08:36):

That's the easy miracle.

Stefan (08:37):

Yeah.

Ilia (08:37):

No problem.

Stefan (08:38):

So yeah, coming to a city near you... What's a good name for this? Mega arm lift? Maybe some sort of workout that you do while lifting yourself with your arms?

Ilia (08:49):

Octopus.AI?

Stefan (08:53):

So Octopus.AI is a new company that is coming to a city near you where anyone with a hundred dollars to spare and a good eight minutes to wait can have a robotic arm pull up to them and get them over whatever is blocking them, whether it's stairs or just a bad time. Octopus.AI now raising money on Kickstarter, entry price - a million dollars.

Ilia (09:17):

Tagline is, "It's a sucker for you".

Stefan (09:21):

All right. That's a good look. That's good. I'm pretty happy about how that went. Cool. So for this next section, I was hoping we could change it up a little bit and have a bit of fun while exploring why some projects that we've come across recently have been harder than the smart and people Biden have thought. And essentially for this next section, if you're listening to this and you think it's about you, it might be kind of about you, but frankly it's about several other people at the same time. Because we've had a bunch of different folks we've talked to in the last few months have really well-meaning interesting projects and then they've kind of gotten stuck with normal robotic stuff, which is interesting because to the number, every single one of these people's been really smart. As our example version of this, I'm just going to say maybe we're looking at automating an oil field and I'm pointing out an oil field because we don't actually have any oil food customers.

(10:16):

We're interested in having some oil field customer with, if you want automate oil fields, come talk to us. But I want to make, I'm going to talk about that industry because we have not gotten really far with anyone in it so far. But let's say, hey, the price of oils going up a whole lot. Shortage of skilled labor and in oil producing areas is a challenge. Maybe it's of a national, maybe it's some big interest of a government or a region or whatever to solve systemic labor shortages in oil field. And the way to do that, the way that we're interested in doing that is we want to make an autonomous oil field, whatever that means.

Ilia (10:52):

Probably the prospecting and drilling operations and-

Stefan (10:54):

Not even prospector, just like, we have some rings that are drilled.

Ilia (10:59):

Oh, okay. You need to service them?

Stefan (11:01):

We need to bring stuff with them. We need to do oil stuff with the new royal derricks and we can't have people do it. So, let's say we have a big budget, a big government budget that feels monstrous for which just needs to be a bit. Doesn't kind of work just a POC. And we have in the range of five to 50 million dollars. So, Ilia, how hard can it be? Where do I start off with?

Ilia (11:22):

Well, given my background, I'm going to start thinking about the robot itself. But the funny thing we notice-

Stefan (11:26):

There's lots of things that we need to do.

Ilia (11:28):

Yeah, exactly. The funny thing we've noticed is that the environment itself is really what's going to be tripping you up to start with.

Stefan (11:35):

How so?

Ilia (11:36):

Well, a lot of these oil fields are not going to be in heavily populated areas. There's a few, but generally the ones, like L.A...

Stefan (11:36):

Famously.

Ilia (11:42):

Exactly. But generally the ones that would be new fields that you need to develop are going to be in the middle of nowhere. So you have no internet infrastructure. So, Starlink is starting to help with that right now.

Stefan (11:54):

Well why doesn't Starlink just solve that problem? I mean, it's single-handedly has beaten Putin.

Ilia (11:59):

Starting to help with that. But so far, we've found in our testing that the downlink rate is great, but the uplink is pretty terrible.

Stefan (12:07):

Well what does that mean? I mean you just have to download how the thing drives. Why do you need uplink?

Ilia (12:12):

Yeah. So for our use cases, what you really want is you want the robot to be as autonomous as possible, but that still means you likely want to have a human driver be able to take over. And when the human driver takes over, they need to see what the vehicle sees.

Stefan (12:25):

But they don't need to see that much. I mean it's just...

Ilia (12:28):

Even a camera is a lot of data.

Stefan (12:31):

How much?

Ilia (12:32):

We pretty routinely saturated 10 megabit per second connection with a few camera feeds. And at home, if everybody's been using Zoom quite a bit these last few years, a Zoom session can saturate a crappy home internet pretty easily. And imagine running a few of those in parallel in high quality uncompressed video.

Stefan (12:52):

Whoa, listen. This is of grave importance to us. We're homies with the cellphone companies. They will install a five 5G tower for us. So we have connectivity now. So, it's solved.

Ilia (13:06):

Yeah, so that's great. And that takes a bunch of the problem. The second problem now is whatever equipment you're building, are you thinking you're building fresh equipment or retrofitting existing equipment?

Stefan (13:17):

Well, I mean the thing is while we have five to 50 million dollars, it's like we do have a budget. We'd rather spend less money.

Ilia (13:24):

Yeah, so retrofit existing stuff probably?

Stefan (13:27):

I mean I guess it depends on how much is the retrofit going to cost.

Ilia (13:30):

So, there's a lot in existing equipment that's been thought through that you don't want to build from scratch.

Stefan (13:36):

Okay.

Ilia (13:36):

Right? If you have a truck, whoever made the truck has already thought of the correctly sized engine, the right suspension, the right tires, assembly processes, all that stuff. So retrofitting-

Stefan (13:45):

Basically we can stand on the shoulders of the giants.

Ilia (13:48):

Exactly.

Stefan (13:48):

Figuring out how to move tools from one side of an oil field to another.

Ilia (13:51):

Exactly. So retrofitting is almost always going to be faster and cheaper, especially for a demo. But retrofitting has its challenges. One thing we've run into and a lot of our customers have run into is that vehicles built for humans assume that the human is going to turn it on and off.

Stefan (14:06):

Yeah.

Ilia (14:07):

And so, something-

Stefan (14:07):

I mean that's just a little signal.

Ilia (14:10):

A little signal.

Stefan (14:11):

How hard... I mean why don't you just put an actuator on the ignition and then it turns on and off?

Ilia (14:18):

Yeah, I mean that's functionally what we did. We have a little relay that powers on the vehicles, but the problem is that relay has to listen for a signal at all times.

Stefan (14:27):

So I'm [inaudible 00:14:28].

Ilia (14:28):

It's surprising, you know?

Stefan (14:31):

I don't want to call out here while asking my leading questions of Ilia. This is a stupid problem that I just had to sensor myself for. This is a ridiculous problem and a problem that I would say is indicative of the broader state of in production robots in the world. We have looked for something like this that is open source that we can just use because we do not want to build any hardware and yet we could not find it. So, Ilia, how many versions of this have we built?

Ilia (15:05):

I have probably four or five by now.

Stefan (15:06):

Fantastic. And how many of them are we selling these to customers go or is this our go-to market model?

Ilia (15:14):

No, this is all horrible, built in my basement kind of... Horrible, horrible tech.

Stefan (15:20):

And if you were listening to this and you were interested in getting in the ignition over internet protocol business, we would love to open source what we have built. We would love to pay you a hundred to a thousand dollars a unit to make these for us so we can tell customers to just buy them from you.

Ilia (15:37):

So here's the challenges. Let me lay out the engineering challenge for any listener who does want to tackle this. You want a device that is roughly the size of a cell phone. Let's call it that size, that can handle anywhere from 40 to 200 amps, peak, because they need to crank the engine, that needs to run off a small battery that has to be able to run from anywhere from minus 40 Celsius to plus 40 Celsius.

Stefan (16:00):

Normal range.

Ilia (16:01):

Yeah, like industrial range, right? And the thing has to be able to sit quietly and wait for an activation signal for a year or two just in the field somewhere and within five minutes of a signal power on the rest of the-

Stefan (16:14):

But, Ilia, why do we need that? Why can't we just have a guy who turns the ignition?

Ilia (16:19):

Yeah, but then you've eliminated the benefits of your autonomous industry in a lot of ways, right? If you need a guy-

Stefan (16:24):

That guy is kind of hard to get.

Ilia (16:26):

Yeah. Yeah. So, exactly, right? So if it's oil fields in middle nowhere, now you're talking about a guy who's going to drive out two and a half hours to turn a key to then drive it back.

Stefan (16:37):

I think he just works there. Maybe he's a full-time robot minder and every day at 9:00 AM he goes down the row of vehicles and hits the ignition on all of them.

Ilia (16:45):

So, there are a few companies that are pursuing this kind of business model where they have X amount of people per Y amount of robots, hopefully one to 10 or something, whatever the number is. If you do the math on that business model, you notice it's drastically less appealing than just pure robots.

Stefan (17:02):

Yes.

Ilia (17:02):

Needing to have a human involved in anything means your costs and your time response are just going to...

Stefan (17:08):

Okay. So for our autonomous oil field, we've figured out how to turn the engine on.

Ilia (17:13):

Yeah.

Stefan (17:13):

We've figured out how to talk to the robot over the internet. Good to go, right?

Ilia (17:18):

Yeah. So, that's table states.

Stefan (17:21):

And to be clear, when we start these tests, this is a pretty valuable oil field. It produces, I don't know, 3 million dollars a day of oil. I don't really know how much oil is worth. One oil. It produces one oil a day. That is great. Nice dinosaur juice for my SUV. If you are going to test here, you can't crash into stuff. That's just cool and easy.

Ilia (17:44):

Yeah. People who want robots to work perfectly off the bat.

Stefan (17:49):

Yeah.

Ilia (17:50):

That's an interesting one where they say, "Yeah, please come test, but, by the way, if you hit any of this equipment, it's 10 million of downtime." So, operators are very nervously watching their vehicle.

Stefan (18:00):

Which quickly eats through your 50 million dollar massive budget. That is so exciting. And even if it's not an oil field where you can bump into something, if it is a valuable physical asset like a farm or a port or whatever, taking it out of commission for a day is hundreds of thousands if not millions of dollars.

Ilia (18:20):

So, the way this goes is your first deployment, you're still somewhat shaky on how everything's working, your hardware might be working, it might not be working. You're kind of doing your initial testing and you need to essentially shut down operations. At least a portion of your area has to be dedicated to your testing. And on oil fields, I don't know how that is, but for something like a mine, that's almost a no-go. Right? Shutting down a mine for any amount of time is extremely expensive. And so in your test planning when you're developing these solutions, you have to think about how am I going to test an unproven vehicle in a live mine with actual operations going on?

Stefan (18:57):

I've actually talked to somebody who they could only look at their hardware for 15 minute increments every third day because that's the only time that the vehicle came to the surface of said mine and that they could take it out of commission at all. But it had to be back to working shape within 15 minutes.

Ilia (19:14):

So, plan out having bad internet access and very little access to the actual vehicle to troubleshoot when something goes wrong. One of the things we've been doing that I highly recommend to everyone is to make sure that any retrofits you're doing in the vehicle do not remove the ability for a human to drive it so that a human at any time at the vehicle just stops and is having a critical error, human can jump in and drive away and keep doing functional work because having it completely dead can-

Stefan (19:41):

Another separate mining example that I've heard about this is in certain flavors of coal mines, if the wrong vehicle shuts down in the wrong spot, it costs the site 10 to 50 dollars a day top line. In other use cases, the wrong vehicle shuts down in the wrong place, it means the next cargo dealer ship can't be filled on time, which means the processing plant, to see an ocean away, go runs empty for a day or two, it means the whole company isn't profitable that month.

Ilia (20:06):

Yeah. It's very, very, tight, tight margins. But let's say your oilfield isn't that bad. So what you really want to do is do as much testing beforehand, offsite as you possibly can. Don't leave any of your testing. You have to be sure by the time you get to the oil field that everything's one hundred percent.

Stefan (20:24):

But listen, I mean with our 25 million budget, we can't exactly go buy you 20 million dollars of equipment to play around.

Ilia (20:31):

Yeah, exactly. So simulation starts to play a serious role.

Stefan (20:35):

But is that simulation super high fidelity where every sensor thing is?

Ilia (20:40):

I find that kind of simulation, I gave a talk on that recently and I find that kind of simulation has real diminishing returns.

Stefan (20:46):

Why is that?

Ilia (20:47):

Because the more effort you're putting into your simulation, the more you're trying to get an accurate model of the world, the more you start to depend on it. And your sim will never reflect every weird situation in the actual world. The example I gave is there's a lot of simulation effort that's been put for autonomous vehicles, but then somebody took a video of a Tesla following a highway and the Tesla autopilot system shows traffic lights flying towards the vehicle and you think for a second, what the heck is going on? And then the camera pans up and it shows that the Tesla is following a truck that is carrying stop lights to be installed somewhere. And so you can't plan for that in simulation. You can't possibly simulate every weird, we call them, kind of, black swan events. And so you're all always going to have to test in real world to verify the performance of your sim. If that's the case, if you have to test in real world, the more time you spend in sim, arguably the less time you're going to spend on real world. So you want to get your sim robust enough to be useful and to get over the hump of, I can't test on real vehicles. But you don't want to spend so much time that you're killing your real world testing time.

Stefan (21:56):

Another version of this that I've seen is given how complicated these systems are, you know have the business logic that says what to do, you have the autonomy that tries to drive there and you have the robot that is actually moving. The more complicated these systems are, the more dependent on each other they become, the more everything fails if anything fails. So a version of this that we've seen is somebody who kind of fully builds everything in house, and as a result, sometimes they couldn't get a field to test and sometimes they can get a robot to test in the field. Sometimes the hardware wasn't working, sometimes the autonomy wasn't working, sometimes the business logic wasn't working. And if any of those things didn't work, which is to say almost all the time, then nothing could actually be tested.

Ilia (22:37):

Exactly.

Stefan (22:37):

Whereas when we started being able to help those tests by putting the autonomy portion in sim, the business logic could rapidly improve. The hardware, because it no longer could say, oh the untested autotomies was breaking me, could actually improve to a point where it was usable. Which then meant that the fuel type that was provided wasn't spent with people sitting in a hot car trying to debug things on the business logic, but more it was spent with the vehicles actually driving back and forth.

Ilia (23:06):

And to make sure that that actually happens, you have to, there's a few kind of ground rules that you have to do with sim. So one of them is that the code you're running in sim, in the ideal case has to be identical to what you're running on the vehicle. We try to push for that as much as possible. There's going to be small differences. For example, simulated sensors and real sensors behave identically, and you know-

Stefan (23:26):

But you separate out your business layer from your, that doesn't really matter.

Ilia (23:29):

It doesn't, really.

Stefan (23:30):

The fact that you can't perfectly simulate a brand new LiDar in a weird environment doesn't really matter if the simulation is more for the point of proving that ordering this tool carrier to go from the maintenance shed to oil generic number seven to see that that actually works every time that the command gets spout out the right way, the vehicle is able to drive.

Ilia (23:49):

Exactly. And if you can build your sim such that it reproduces at least roughly your actual use case, I'm talking about GPS coordinates, terrain features, distances, real kinematics of your vehicle, then when your business logic tells the vehicle to go, the estimate the vehicle provides back in terms of ETA will be roughly accurate to the real estimate. So you can use that to really test out how good is your UI. Does it make sense in terms of how responsive it is? Because another trap that people fall into is when they build their simulator, they take a lot of shortcuts, so that when you send a command, it immediately replies back saying, "Yes, I'm there". And those sorts of traps are avoidable if you build your sim to be run the same code as your real vehicle.

Stefan (24:31):

So, we have our oil field, we have the ignition, we have data. We've now tested out the business logic of tool carrier to Derrick seven and did it seems to go, what do we need from there?

Ilia (24:47):

So part of simulation step is to actually ensure that you know what sensors you're placing and where on your vehicle, especially for a retrofit. So during that simulation step, you probably have gone through a few iterations of how many LiDars do we need? Cameras, radar, ultrasound, whatever it is for your particular use case. Then there's the step of actually installing that on the vehicle. Again, in sim, you're just going to have your sensors sort of magically floating in space. In the real world, you have some mechanical effort to build mounts, do wiring, make sure that your computers are communicating well with the sensors, make sure that there's failover, that sort of stuff. There's also actually communication with the vehicle itself. If the vehicle can provide back any kind of odometry on speed or engine health or fuel consumption, all of that should be taken to account if possible because then you can make more intelligent decisions.

(25:34):

You can report back to business layer that this particular vehicle is low on fuel or is overheating or whatever. That full integration step is extremely manual and is very painful and is full of. I accidentally stripped a wire too short and now I have to add another wire. Or this connector doesn't fit properly or this casing cracked. So again, as much as you can of doing that offsite and doing that before you get to the oil derrick, the happier going to be because getting spare parts out to these test sites tends to be a week long effort and having it right near your workshop-

Stefan (26:11):

And how popular is it when the royal derrick shut down for an extra week because you're waiting for spare parts?

Ilia (26:16):

Especially if we're waiting for a 5 cent widget you just don't have on hand. That's why a lot of these sites will run on duct tape and chewing gum because you just can't afford to get the correct parts to fix it. So get out all those gremlins as much as you can ahead of time and still, even if you're completely one hundred percent sure that it's working, bring a spare of every critical part. You just really don't want to wait.

Stefan (26:40):

All right, so now you know where the sensors are supposed to be acquainted. They've driven in your test field, you've tested the business logic in sim, you have connectivity, the thing can start itself. What's next?

Ilia (26:54):

So then it's commissioning. Then it's basically, you've tested offsite that your control is working well, now you're onsite. You need to verify that all your simulation and all your offsite testing is working.

Stefan (27:05):

So in other words, that's when the technology actually starts.

Ilia (27:08):

That's when the real test starts. Yeah, all this effort has been build up to actually starting to test.

Stefan (27:17):

Which I think is kind of the point. And what's been frustrating about these flavors of conversations that we've had is we've talked to really smart people who mostly are just like, yeah, sure. Just come and test. Yeah, sure. We don't need to see a perfect thing. Just show us an early test. And the challenge, as it's so often is in robotics, is that all of the real work is before that early test.

Ilia (27:40):

Yeah. And can you skip these steps and get lucky on site? Absolutely. I've seen the happen with teams, but even the really lucky teams, what the test day looks like is four to six hours of head scratching and soldering and debugging and 10 minutes of testing. And then maybe you're happy with that.

Stefan (28:02):

And that 10 minutes of testing is a very high risk 10 minutes to try to get an investor, a customer, a partner, or whoever to watch it.

Ilia (28:09):

Exactly. And that will happen. And if you do do all this prep work, then your actual testing time can be four or five hours of productive data gathering to double check your assumptions to actually go back later and actually verify that the models you've built and the control algorithms you've built perform as well as you think they do to gather metrics on actual use case performance and not just simulated performance.

Stefan (28:37):

Which is of course what you need to get to the point that you can actually deploy.

Ilia (28:41):

Yeah. Both of these are a demo. One of them is a demo you can build on. The other one is demo you have to throw out and start again.

Stefan (28:46):

And frankly, the demo that you can build on is something that you can, as a business person feel comfortable in inviting some outsider to. We've, I've watched tests like this in many flavors of life where an investor comes out and they spend the whole day in their car looking at their phone and disappointed and confused as to why they can't see the robot just work.

Ilia (29:08):

The really fun part about all this is if you do everything absolutely right and it works well, every single person other than your team will be like, yeah, of course. What did you expect? Of course it's going to work well.

Stefan (29:18):

How hard can it be? Robots have been around for a while.

Ilia (29:21):

It'll impress literally no one except maybe your parents. Everyone else will be like, yeah, okay, whatever. So, it worked well.

Stefan (29:29):

And on that note, yeah, I think this was a fun one of how do we actually do something?

Ilia (29:35):

Yeah, there's a lot more complexity than you'd think.

Stefan (29:39):

All of the challenge as it is in robots comes after you do all the right work. Of course at Polymath, we'd love to help you get to the good successful test sooner rather than later. Hopefully, I think some people can automate some more lyrics and other associated structures.

Ilia (29:55):

Absolutely. Alrighty, that sounds good.

Stefan (29:57):

Well with that, we'll catch you all next week and we'll talk about some more robots to automate.

Ilia (30:02):

Talk soon.

‍