Oct 22, 2026 · 46 min · 11 segments
In this episode of *Future Bytes*, host Magnus Oxenwaldt sits down with Kacper Nowicki, co-founder of Nomagic, to explore what it really takes to turn robotics from an impressive demo into a reliable…
Kacper NowickiGuest
Magnus OxenwaldtHost
You know, one of the things that we've started doing very early in the company is that, you know, what we are selling to the customers is a service and it's not a machine or technology, right? So I think this really makes the adoption and relationship in the right place where, you know, when we sign a contract, It's a contract that states the availability of the robot throughput of the robot and autonomy of the robot.
So something that the customer can really relate to and they don't have to worry about a brand of a component or a lot of details.
Because if we agree on the contract level on something that you know, their business really needs, and then we deliver this, then, you know, the whole technology stack is on our side, right? So I think this ensures that in the end, the customer gets what in the end their business needs.
Now, on the technology side, and I'm a technologist, so there's a lot of exciting things that we do to be able to deliver this super high reliability.
But maybe before I talk about this, I would talk about the mindset, right? Yes.
i'm an optimist you know i believe that uh this change that we're bringing to the world is is going to have a good impact and and you know we'll be able to scale to thousands of robots soon but i'm also a very practical person and uh what has what has uh what we've done in a company is that you know we very carefully look at every failure every malfunction of the mistake that robot does right And being positive about the overall outcome, I would say we're a little paranoid about every little mistake.
And this helps us to really understand, okay, what is stopping the system from giving this fantastic value to the customer? How can we improve it? And what are the...
kind of technical solutions that are needed to address this right and looking closely at the robot performance uh in production right like brings a lot of interesting facts that you uncover and then you need to somehow deal with this right and one fact that uh We've learned that there are these physical world, I don't know, mistakes or phenomena, right? I was looking at the statistics some time ago and on a day a few months ago, we had eight cases when the item we're picking was glued to another item, right? And we pick one little white box and there is another box glued to this.
And then this box either is delivered or the first box was delivered or is lost on the way.
when they're sent to the robot, but the customer doesn't know, right? And there are these cases.
And now we need to somehow handle them, understand them, and not ship to white boxes to the target customer, right? So understanding that this is happening, you can then build a solution and a technology you know, solves this problem.
But there are really many cases like this of unusual behavior, which, you know, once you have, you know, millions of peaks a month, right, like in production, then you start to see these these kind of rare cases and then you find a technical solution to this.
And I'll just mention that this collection of what happens In production, you know, we have a kind of a nice name for this library of chaos, right? Like, you know, all the weird things that happened, you know, in production that we catalog and we understand and then we address, right?
Can you learn something from this library in terms of, for example, the more you generalize the behavior on a robot, let's say it has a very narrow spectra of tasks it's about to do, but the smaller the business we want to spend on, the more we need to generalize because we might not have the volume of task, right? Do you see like a strong relationship between the library of chaos and generalization where the more you generalize, the more chaos you will actually see in your library?
Yes, I think there is something like this, right? So our robots learn from the fleet of all the robots in a field, right? So even if you have a maybe smaller business and 1 robot, right? Like, it's benefiting from another 100 robots that are out there.
And and I think even if you haven't, you know, we haven't seen good items in your operations yet, but for whatever reason they happen in the future, then the system is already aware of this type of behavior.
Like our machine learning systems are trained on the whole library of chaos.
The interesting thing about the systems that are based on machine learning is that you need to really create a distribution of different types that happen for them to learn appropriately.
If you give them 99.9 percent successes and only 0.1 percent of these failures, then the system will ignore them because for it to optimize the behavior, it's enough to focus on 99.9.
You know, one of the things that we've started doing very early in the company is that, you know, what we are selling to the customers is a service and it's not a machine or technology, right? So I think this really makes the adoption and relationship in the right place where, you know, when we sign a contract, It's a contract that states the availability of the robot throughput of the robot and autonomy of the robot.
So something that the customer can really relate to and they don't have to worry about a brand of a component or a lot of details.
Because if we agree on the contract level on something that you know, their business really needs, and then we deliver this, then, you know, the whole technology stack is on our side, right? So I think this ensures that in the end, the customer gets what in the end their business needs.
Now, on the technology side, and I'm a technologist, so there's a lot of exciting things that we do to be able to deliver this super high reliability.
But maybe before I talk about this, I would talk about the mindset, right? Yes.
i'm an optimist you know i believe that uh this change that we're bringing to the world is is going to have a good impact and and you know we'll be able to scale to thousands of robots soon but i'm also a very practical person and uh what has what has uh what we've done in a company is that you know we very carefully look at every failure every malfunction of the mistake that robot does right And being positive about the overall outcome, I would say we're a little paranoid about every little mistake.
And this helps us to really understand, okay, what is stopping the system from giving this fantastic value to the customer? How can we improve it? And what are the...
kind of technical solutions that are needed to address this right and looking closely at the robot performance uh in production right like brings a lot of interesting facts that you uncover and then you need to somehow deal with this right and one fact that uh We've learned that there are these physical world, I don't know, mistakes or phenomena, right? I was looking at the statistics some time ago and on a day a few months ago, we had eight cases when the item we're picking was glued to another item, right? And we pick one little white box and there is another box glued to this.
And then this box either is delivered or the first box was delivered or is lost on the way.
when they're sent to the robot, but the customer doesn't know, right? And there are these cases.
And now we need to somehow handle them, understand them, and not ship to white boxes to the target customer, right? So understanding that this is happening, you can then build a solution and a technology you know, solves this problem.
But there are really many cases like this of unusual behavior, which, you know, once you have, you know, millions of peaks a month, right, like in production, then you start to see these these kind of rare cases and then you find a technical solution to this.
And I'll just mention that this collection of what happens In production, you know, we have a kind of a nice name for this library of chaos, right? Like, you know, all the weird things that happened, you know, in production that we catalog and we understand and then we address, right?
Can you learn something from this library in terms of, for example, the more you generalize the behavior on a robot, let's say it has a very narrow spectra of tasks it's about to do, but the smaller the business we want to spend on, the more we need to generalize because we might not have the volume of task, right? Do you see like a strong relationship between the library of chaos and generalization where the more you generalize, the more chaos you will actually see in your library?
Yes, I think there is something like this, right? So our robots learn from the fleet of all the robots in a field, right? So even if you have a maybe smaller business and 1 robot, right? Like, it's benefiting from another 100 robots that are out there.
And and I think even if you haven't, you know, we haven't seen good items in your operations yet, but for whatever reason they happen in the future, then the system is already aware of this type of behavior.
Like our machine learning systems are trained on the whole library of chaos.
The interesting thing about the systems that are based on machine learning is that you need to really create a distribution of different types that happen for them to learn appropriately.
If you give them 99.9 percent successes and only 0.1 percent of these failures, then the system will ignore them because for it to optimize the behavior, it's enough to focus on 99.9.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
