Mytra’s Ahmad Baitalmal on Aligning AI and Robotics with Physical Constraints

ai video series. I'm your host, Mike Za. Today we're talking with Ahad Al, who, CTO for Mitric, and we're talking about how things like warehouses are being automated using various robotic platforms and software and all kinds of fun stuff, but maybe we need to actually physically understand what it is we're trying to automate first, because, well, a lot of people are putting the cart before the horse.

Ahad, welcome the show. Thank you. Great to be here a lot.

What is happening here with these types of projects? Because on the one hand, I wonder, do we really wanna replicate what we already have and is that what's working? And or people trying to jump ahead and re-engineering the entire thing without understanding what it is they're re-engineering.

And so maybe we do need some more hands-on experience with the process. I, I, I think there's a, um, an approach that we've seen before where people see a new emerging market that is, uh, interesting and, um, uh, they see an opportunity to create value. And so they jump in.

Um, when, uh, social media started, uh, taking, taking off, uh, you have a lot of entrance to say, Hey, it'd be really nice to have, you know, a social media platform. And a lot of apps be built trying to build that. Um, many of them failed.

Uh, some of them, uh, one may be by mistake, but a lot of the, uh, uh, the right approaches is to get customer insight. Meaning try to understand what you know, what your potential customers are trying to solve first. And that is hard to do by guessing, uh, especially from the outside.

If you haven't experienced, uh, when yourself, uh, if you take, uh, Instagram as an example, you know, people might have an idea about, I just wanna share pictures. Great. See if, if it's a workflow that, that you actually enjoy yourself first and go, I wanna share pictures with friends and, you know, become a customer of that service first and see what works, and then go ahead and say, okay, now I know what to do.

I'm gonna go build an app that does something special. This is the same thing now applied in industrial robotics. It's really a hot market right now.

Uh, a lot of entrants are coming in and, uh, it's easy to go and say, I'm gonna build this nice looking, you know, uh, OID robot, that kind of baby dances. And, uh, it looks great. Wonderful.

It's fun. I want to do that as an engineer. Wonderful.

But then you have become a customer first. The reason, uh, we believe, you know, our approach is, um, uh, more applicable is we've been customers of material flow at industrial, um, uh, manufacturing, you know, pe rivian. And it wasn't that we wanted something that looked nice or really cool, we had a real problem to solve.

And we thought, oh, what really is important right now as a customer, what it would solve? My problem is a, an efficient flow of material that was, uh, easy to deploy, that had, uh, minimum number of integrations and complications. After that, after that insight, I know what I, what would make my life great.

I go and invest and build a new solution that does exactly that. And so the step of you actually doing, uh, becoming a customer is generally skipped. People go, I know what to do.

Yeah, this would be great. You're making assumptions. And it's not a matter of, um, serving people.

Say, Hey, tell me what's important. It's kind of a, uh, uh, uh, known wisdom that the customers will never tell you exactly what they want. They, they'll tell you like, I, I need a faster horse.

Like, maybe it's a car. Mm-hmm. And so you will not get that insight by just asking, you have to physically become the customer.

Uh, in the case of warehouse or industrial, um, or manufacturing, go to the floor. I mean, it's not really hard. Go to the floor and say, okay, show me how it works.

Become that person, work, uh, in, in the factory, work in the warehouse, and see what are the friction points that with your skills as an engineer or whatever field that you're in, how can I solve this, uh, from my, from my perspective, um, that, that is the approach that, that we're pushing. And as I understand it, it's not all that simple because a lot of the AI technologies that we're trying to embed into these industrial robots don't really understand physics just yet. And so they don't understand the movement and how much distance there is between different things and how much hitting something might hurt.

And a lot of times a lot of these robotic projects get scrapped, and that costs millions of dollars simply because they didn't think through the physics of the thing. So, um, what is your sense of, uh, what do we have to overcome here to kind of make the industrial robot smart enough to see and understand what it's looking at? Uh, it, it needs to, um, there are a few data points that need to be collected more than just saying, uh, I have a great, uh, generative ai.

It can answer a few questions. Now I'm gonna take that and apply it to, uh, physical world. And it's just gonna look, uh, maybe you really have to try it out and be ready to, uh, admit to the gap.

Say, Hey, there is a gap here. I invested all this time. There's a gap between the degenerative part or the thing that the AI is giving me, and what really needs to happen.

Um, it could be that you have some wishful thinking, like, it, it really must work. It's really cool. I, it'll, it'll actually work, but unless it is, um, uh, actually effective at, uh, giving you more productivity, you need to pivot.

Uh, this is, you know, pivoting is a very, uh, well understood, uh, business process where people go, Hey, this, this didn't work. You shouldn't be emotionally attached to it. This did not work.

Let's move on to the next thing. Or say, Hey, it needs more tweaking. Uh, get it done better.

Uh, I think there's a little bit of wishful thinking, um, that, that comes with the deployments of AI that says, this will solve everything. It, it's smart enough that that's it, we're done. We don't have to invent anything new.

Um, it that, that has yet to be proven. Um, it is not impossible, but I think the homework hasn't been done yet. Deploy the ai, see what the gap is, try to fill it, uh, try to fill to, you know, bridge this gap.

Uh, that doesn't happen by just, um, building an algorithm, handing it over as like, Hey, this will solve everything you have to actually apply. So where would I find the people who understand the warehouse or wherever I'm gonna put this industrial robot and have the AI skills? I mean, am I looking for a unicorn or is there, is this more of a team sport?

How does that come together? We're lucky in the, in the Bay area, we have, um, multiple disciplines, uh, that come together that, uh, you know, feed each other and, um, uh, tech and software is, is prevalent in this area. And so when manufacturing started having California with Tesla, uh, it was a fresh look, uh, at, uh, you know, established practices.

Uh, I think the best, uh, uh, people, uh, and, um, uh, at least culture approach to solving these problems right now is to look at other disciplines, uh, how they solve things and apply it, let's say, uh, industrial automation. The best engineers have a good solid background in the problems, say warehousing, uh, manufacturing, uh, uh, industrial automation, and they start to branch out. And like, how was this problem solved in, uh, computer science?

So engineering or, uh, DevOps where you take that, uh, uh, innovation and apply to the physical. A very good example is, um, if you notice, uh, maybe 10, 15 years ago, we started not worrying about the search engine, not not showing up or your mail not arriving because, you know, companies like Google and then other providers that kind of figured out how to get a service to reliably, uh, be available, extreme high availability, that's their business. Um, they had to invent new processes, you know, things like pre containerization and Kubernetes and, you know, fault tolerate design.

Logically, those things were solved. Today, we don't really worry too much about opening your browser and, and getting a search result. It's, it's extremely reliable.

And that was a, a, uh, an innovation in distributed computing and reliability that happened in that discipline, um, that can easily migrate into the physical world. How do you ensure that a production line runs exactly like a stream of, you know, data, uh, is always available while you build in this fault tolerance? When one machine dies, the next machine takes over automatically and smoothly.

Uh, the data is shared between machines. It's a stupid extraction. These disciplines, um, uh, or innovations haven't migrated to the physical world, uh, fully.

And there's a huge opportunity to apply these, uh, these techniques to the physical world where you can look at your material flow, just like data running in a, in a data center. Um, if you apply the same principles, uh, to the physical world, the people who understand how this physical world operates today, um, uh, indoctrinated in, in a new way to do, you know, quick sorting, uh, uh, fault tolerance, uh, high availability, take that and apply to the physical world, and they're the best people to, uh, implement that as we go down this path, am I gonna see kind of these humanoid robots capable of performing multiple tasks, or is it more likely we're gonna see, you know, a network of highly specialized robotic systems, devices, whatever it may be, that are working in concert with each other to accomplish a task? 'cause it seems to me it's difficult to teach something that's humanoid multiple job functions.

And so maybe we're better off just having a little, a lot of specialists things. There is a, um, hypothesis that a generalized human humanoid, uh, is exactly what we need forever. Uh, maybe it's a hypothesis, I believe, uh, at this point.

Um, a good example is the, the bot behind me. Uh, the problem we're trying to solve is, um, material flow in manufacturing. And most of what you're doing is moving the materials, uh, is my direct, um, approach to solving this problem a human, uh, I don't believe it is.

So we thought, okay, what do I really need to do? And I, again, I don't wanna be attached to, oh, humanoid is really great. It might be in the future, but what am I trying to solve today?

For me right now, I want to solve material flow, uh, an efficient way that, uh, increases safety and productivity. So you build a form factor without having any attachments to some preconceived notion of what a solution is. We thought, Hey, uh, we're trying to move 3000 pounds backwards, forwards, up and up and down, uh, left and right.

Uh, is it a humanoid? Forget about that. Now let's, let's focus on what the form factor needs to look like.

The form factor looks like a thing that carries a pallet that moves 3000 pounds. And, uh, today, forklifts that do this are 9,000 pounds at least, and they cost 10, 20, $70,000 each. And we're trying to say, what am I trying to do?

I'm just trying to move this material in these directions. And so to do that, I could come up with a device that is 500, 600 pounds only, and it'll do exactly the same work or a fraction of the, the energy, and it increase my productivity in the end, I solved the problem. I did not, you know, get distracted by, I need to build a humanoid, uh, maybe a humanoid would've solved from, but really, the way that you approach from first principles is, here's the thing that needs to be solved.

What is the best form factor for it? It's almost like biology. Uh, the best evolved thing that extremely is, that is extremely effective at solving this problem, is the thing that wins.

And this is the form factor that, that work for us. I believe that companies that want to innovate and improve in, um, uh, industrial automation need to look at the problem and, uh, work backwards. Not say, I know the solution, I'm just gonna go build it.

No, go, go and invest and see what are you trying to solve? What are the first principles, uh, that lead you to arrive, uh, to the best solution? And you come down to the numbers.

I'm trying to move 3000 pounds and three, uh, 3D the best form factor for that looks like what you see here. Um, it has the elements of the humanoid, it balances it, it has a lot of technology that we invested there to, to accomplish that. The climbing up and down, um, uh, from any cell to any cell, it's a big Minecraft grid.

It is innovative, it is very cool. It just didn't have, um, uh, the preconceived notion of, oh, it's gonna look like a humanoid. If I were to build a humanoid today, um, to solve this problem, I would ask the humanoid to get on a forklift and stop moving.

It's like, okay, so I didn't solve the really the problem. Um, I'm trying to move the material efficiently and, uh, work from there. And you arrive at something different and innovative and perfect, uh, for the problem that you're trying to solve.

How long will it be before we achieve this great future? I mean, I know that we're seeing robots today in factories and warehouses, but this notion of a robot that is more, uh, cognizant, shall we say, um, where are we on this term? Um, it is hard to tell because we've been promised, uh, if you remember back in the eighties, maybe late seventies when automation started, uh, appearing in, uh, uh, auto manufacturing at that time, people were like, oh, it's over.

That's it. Robots are taking over. Everybody's gonna be, you know, uh, out of a job nobody's gonna build anymore.

Uh, and I was get, and I was gonna get a flying car that I didn't have a job to pay for. Exactly. Uh, so, and you see, until today, I mean, uh, very little automation is practically deployed.

It's, uh, it's still a manual, uh, world. And so there is a thing that just, um, uh, prevented that from happening. Um, it was too optimistic to say the robots are just gonna do it.

There's, um, a lot of variants, uh, that robots are not good at, that humans are really good at. If a, um, a box falls while then in the warehouse, if the box falls on the floor and I'm supposed to go pick it up, if it was a robot, like it'll just sit there unless you go as a programmer and program all these exceptions in there and say, Hey, watch out for a falling box. Here's how you solve it.

Watch out for this. Here's how you have to pre-program it to do that. You throw a human at this problem.

You don't even have to tell them they see a box on the floor that shouldn't be that. They'll go pick it up and continue the job. That gap between handling exceptions without needing any training whatsoever.

And, uh, you know, having, uh, prescriptive directed, uh, you know, programming for the bots is, is the thing that's lacking the promise of ai, especially at, um, the influence level at the edge on the bot itself is very interesting. Now, if you have these inference models that runs on these, uh, robots, uh, on the floor without needing any cloud or anything, it's interesting. They start to, uh, handle, handle the exceptions that humans used to do.

Um, what's happening right now is the reliability of it. And we've all seen examples of people asking Chad, GT or any one of the LLMs asking a question, getting a ridiculous answer back, or, that's insane, unfortunately, until that has been solved, where you're getting really reliable, really good, um, uh, answers, uh, it's still gonna be, we're gonna need some humans in there. The robots are not gonna take over.

You can't really rely on them yet. It's not impossible. Uh, and there is progress being made.

It's very hard to tell, like maybe two years ago when, uh, I believe three came out, um, you go like, how long until it like answers everything for us. Here we are, two years later, we're still going like, right? It's still impressive, but it's still like, there's still something missing.

Uh, I just recently heard about, uh, uh, apple turning off their, their news summarizing, uh, ai. This is a big company that invested a lot, and they had to kind of just say, Hey, this is not the, the headlines, it was summarizing as an AI sounded like really off like, it, it wasn't useful. I just turned it off, not, I mean, it's an attempt and we should always encourage people's like, yeah, it didn't work.

Keep going. And so we're at the phase of promising, let's just keep going. How long will it take?

Uh, it's everybody's guys, everybody's working really hard to get to an end. So it could be this year, could be next year. Um, but it's really hard to say when innovation, when a breakthrough is gonna happen.

Uh, a lot of people have been trying to figure out, you know, uh, flight maybe for hundreds of years. Everybody's like, yeah, possible. Yeah.

And mean, you see attempts after attempts until the Wright Brothers breakthrough was made, and now we have aviation. We just couldn't predict exactly when it would happen, what was involved into, um, robots that essentially we're trying to orchestrate, rather than us performing the tasks, we are becoming the managers of the robots. Uh, that is, uh, hopefully the, uh, the short term, uh, outcome that is, that is possible.

Uh, and it's not binary. It's not like a, Hey, everything is automated and we're just sitting there pushing buttons, or we're doing all the work ourselves there. There's a gradual, um, uh, progression where some tasks start to be reliably automated to the point where we go, yeah, this is, this is good.

Uh, so here's an example. One of our deployments, um, um, the, the, uh, set of, um, pallets that need to be pulled out, uh, used to be manually done, where somebody goes driving around with a forklift, pulling out a pallet manually, like spend a whole day, you know, driving to belt and taking it out and putting it in a trailer, um, a system like ours made it. Uh, so it's simple.

You come to the, to the, uh, the edge of the, uh, uh, the installation and you say, here are all the pallet that are like to pull out of the warehouse a order and just enter it on a little iPad. It's just a very simple, very pleasant experience. And then these things, the, the bots go automatically.

They pull the material out in order, and you just pick 'em up one by one from the same location. You don't have to go search for it everywhere. Um, that is a, you know, it, it's, it's a very pleasant, it made it so you're doing the same work.

We still, you know, picking up pallets and putting in the trailer, but it just made your life way easier. It took off a lot of the mundane, uh, monotonous, you know, uh, work that you have to do, and it made your life easier. Uh, at some point it is gonna make it even more easier.

There are other areas where you just make your life simpler. You're still doing the work, but the, the hard work is, is out of the way. And you don't have to spend your time in a harsh environment.

Sometimes these warehouses are like, when it's 20 degrees, you know, you don't wanna spend the whole day and minus 20 degrees. It's just, it's punishing. And so the hard work, okay, let the bots go get the material out for me.

I'll just take it. I now prefer this than, uh, you know, the alternative, which is physically going to. So there are other areas where, where automation is just gonna make our lives slowly easier and easier.

And, uh, the smarter they get, the better force, because it'll be closer to the push button warehouses taking, you know, it's the dream. And, and, and hopefully everybody will be in a better mood when they get home. But let me ask you this last question.

What's the economic impact of all of this? Because are we gonna see more manufacturing move closer to the point where whatever's being made is consumed? And we might not need to, you know, ship goods halfway around the world just because there was cheaper human labor someplace.

We could just have a lot of smaller factories and warehouses closer to the point where things are gonna be used. Um, the, the impact of, at least in my view, the impact of that is, uh, dwarfed by, uh, the possibility of, uh, enabling manufacturing where it wasn't feasible before. If, if you were to say, um, you know, I, I would like to create a, a new, um, auto, uh, establish a new auto manufacturer in California or Kansas or somewhere else, uh, you have to take into account a lot of inputs, a lot of things that you have to build out first.

Automation is a big part of that. Your warehouse operations is a big part of that, and it comes with a lot of costs. And, uh, if, uh, automation makes it so things that were not possible before are now possible, uh, now you decide, yes, I will decide to go and build that auto factory, and you create more jobs, more economic growth, more opportunities, and know more innovation just because automation added that extra component of feasibility.

Where before it was hard and there you have to ship things from everywhere, but now automation made it so, or it's easier. I can actually, uh, um, uh, decide to actually go ahead and build out these, uh, um, warehouses and, and manufacturing facilities. So a lot of the opportunities that were not feasible will become feasible now that automation is introduced.

So I think the impact, um, is gonna be extremely positive. Um, there's gonna be these, um, uh, e uh, ecosystems that say, Hey, there's a new manufacturing facility now it's highly automated. And now because of it's there, all the restaurants in that area are now prosperous.

So all the real estate gets improved. All the, uh, logistics, uh, that's happening. It's bringing more business.

There's huge impact, uh, that is extremely positive that's gonna happen because of automation. All right, folks, she herndon here. The robots are coming.

The question now is how are we all gonna work alongside them? They come up with something that is a better outcome for all concern. Hey, Ahmed, thanks for being on the show.

Thank you very much. Appreciate it. ai video series.

You can find this episode and others on our website. By all means, check them out. Until then, we'll see you next time.