Expert interview: how do humanoid robots get from pilots to safe deployment?

0:09 Roberto D'Elia: We are moving from demonstration pilots towards real deployment. Today we have a system coming from China, the AI is coming probably from the US, and then it's operated in Switzerland. A compromised industrial robot is not simply an IT problem. It can become a physical safety problem. We cannot certify a robot once and simply assume that it remains the same machine for the next five years. Safety is not a component. Safety is a life cycle. Everything starts about the use case. My principle would be outcome-based wherever possible, prescriptive where failure is unacceptable. The next phase of humanoid robotics will require an ecosystem, not a better robot.

0:59 Interviewer: Your Munich keynote is called “No scale without trust”. Why now?

1:05 Roberto D'Elia: I believe we are reaching a very important transition time period about humanoid robotics. Over the last few years, the dominating question was: can technology actually do it? I mean, if you saw a robot walking, it was like a sensation, right? So now we are talking about something totally different: manipulation, object navigation, complex environments and performing economically useful tasks. So we are another stage in humanoid robotics.

1:38 Roberto D'Elia: So increasingly, the answer is yes. It's the right time, it's the right time for our association to step in.

1:47 Roberto D'Elia: We are moving from demonstration pilots towards real deployment. The question changes. It becomes: can we trust systems to perform this task safely, repeatedly, and accountability in a real-world environment? I think this is very important to do that and address that at that time.

2:05 Interviewer: Will the world agree on one set of rules for humanoid safety, or will every region write its own?

2:13 Roberto D'Elia: First of all, I think we should be avoiding to frame the things very simply. So sometimes: Europe is regulatory, you know, America is innovating and China is moving fast. So this is very simplifying the things. The reality is much more nuanced.

2:31 Roberto D'Elia: But it's particularly interesting that these systems do not need to become identical. So this is totally important. The important question is that it's becoming interoperable. Because, you know, today we have a system coming from China, the AI is coming probably from the US, and then it's operated in Switzerland. So you see, the components are much more complex than just defining these three things about governance and regulation.

3:03 Interviewer: Your association is helping shape a safety standard in Switzerland. What are its key building blocks?

3:10 Roberto D'Elia: So actually, to give you an insight about what we are working on: we identified six essential building blocks.

3:19 Roberto D'Elia: One is physical safety, and I think first the robot has to be physically safe around people. This includes collision risk, stability, force, speed, unexpected movement, safety stopping and emergency stop mechanisms.

3:34 Roberto D'Elia: Then we have the functional safety. We also need to understand what happens when something goes wrong. What happens if sensor fails? What happens if connectivity disappears? What happens when actuators behave unexpectedly? The machine needs to move into a predictable, safe state.

3:52 Roberto D'Elia: And then it comes to cybersecurity. Number three: humanoid robots are connected systems. So we need to understand: there will be remote-controlled software updates, AI model updates, and may change sensitive operation data. Cybersecurity becomes part of the physical safety. A compromised industrial robot is not simply an IT problem. It can become a physical safety problem. So it's absolutely essential that the cybersecurity is framed in this way.

4:23 Roberto D'Elia: Then we have number four where, you know, we think AI and its behavior assurance. So it's not sufficient to test what the robot is designed to do. We also need to understand what happens when the robot encounters something that was not anticipated. And this is happening right now in the AI area.

4:41 Roberto D'Elia: Then we have a life cycle assurance. You know, this is particularly important for AI-enabled robots. A robot receives a new software version, perception model, AI updates, may no longer behave exactly like the system originally was designed for. So we have to move beyond the traditional idea of certification as one-time event. You know, when you certify a car, it goes on the street. Then obviously the brakes, they will behave like this, and it was certified like that. Now things are getting much more complex, and we cannot certify a robot once and simply assume that it remains the same machine for the next five years. So it's very important to life-cycle-assure the system.

5:29 Roberto D'Elia: And then, I think one of the most important ones is the clear accountability. Finally, responsibility needs to be clearly allocated. The OEM has a responsibility. We need to acknowledge that. The software and AI provider has responsibility, the integrator has responsibility, and the operator has responsibility, and the company deploying the robot has also his responsibility. So one of the major governance questions will therefore be defining where one responsibility ends and one other begins.

6:01 Roberto D'Elia: And for me, the critical principle is: safety is not a component. Safety is a life cycle. And I think this is very key of this purpose.

6:11 Interviewer: A European manufacturer wants to pilot a humanoid within 12 months. Where should they start?

6:18 Roberto D'Elia: I think they should be starting thinking about, you know, the use case itself. I mean, everything starts about the use case, because you will be never seeing robots working up front without getting into a use case that can be addressed to these humanoids.

6:42 Roberto D'Elia: I'll give you an example. Companies are coming to our association saying: “Roberto, we want to buy a humanoid to see if it works in our warehouse or whatever, in logistics.” Because I'm actually in logistics, so most of the cases are in logistics right now. So they are coming and saying: “Okay, how should we start? We want to buy one.” “Okay, well, there are 140 producers, so you can choose whatever you want to. So the purpose is essential in this case: for what?” And then mostly the companies are saying: “Okay, we want just to test and see how it works, and we want to make some testing.”

7:24 Roberto D'Elia: And then I get critical. I get critical because, you know, it's not just the machine putting there. You connect these machines. And as I was saying previously, when things get connected, then obviously there's already a risk there. So I avoid to tell them to connect it to the system, because then things can occur right away. And then you can't escape to that. It's not about IT, right? Because, you know, interconnecting the system, it's easy, but to be aware about what's going on and how they can influence your existing system, this is very critical.

8:06 Roberto D'Elia: We avoid to give advice to implement and connect these systems for now without having a safety-critical standardization topic. We give some advice. We have frameworks that we are giving to our associates, but still, I think the mindset is still not there: what this means exactly for a company to deploy.

8:28 Interviewer: Are humanoids already doing real work in Europe? Give us an example.

8:34 Roberto D'Elia: Yes, absolutely. So I'll give you a small example, because we saw a few of them. But one simple task: you know, our company, Planzer, they are manufacturing, assembling, for the retail, some packaging stuff.

8:55 Roberto D'Elia: So a producer brings small bottles of, let's say, Omo or some sort of cleaning stuff, but there are big packaging. Now they need to reduce that to a double pack or a multi-pack of products, and this needs to be made by hand. So you unpack a big load, make it smaller loads, you shrink them, you label them, and then you put it in the perspective of a shelf. And then you transport that again out of the factory, out to the retailer. So this is a normal process for the retailer, and Planzer is doing that for the big companies and corporates.

9:42 Roberto D'Elia: So there's a lot of manual work that people and humans cannot do for more than two hours. Okay, so then you have to shift the place, because from a health perspective it's very dangerous. I mean, you have health issues, you are doing more than two hours the same task.

10:03 Roberto D'Elia: And there is a lot of opportunities for robotic companies to multiple usage of a use case for a humanoid, where he can also complete tasks that are much different, and different use cases can be applied for the same humanoid. And I think there is the use case. The real use case is there where you can place different humanoids in different places at different use cases in one single system. And I think this is the most important one.

10:34 Interviewer: Should safety standards be strict, or leave room to innovate?

10:40 Roberto D'Elia: I believe it's a combination of both approaches. My principle would be outcome-based wherever possible, prescriptive where failure is unacceptable. So I think it's a key point.

10:53 Roberto D'Elia: If standards prescribe every technical implementation in detail, there is a real danger of limiting innovation. And I'm an innovator by nature. So humanoid robotics is deploying too quickly for standards to define exactly how every technical problem should be solved. I think it's not possible. A standard should be therefore not telling engineers exactly how to design a humanoid, I mean, a humanoid hand, as an example, but it should be very clear about what happens when that hand unexpectedly makes contact with a person.

11:26 Roberto D'Elia: But emergency stopping, as an example: I think, you know, if you have seen once a humanoid collapse, there are a couple of kilos going down, unexpected, in a direction where you cannot control. And I saw a couple of videos also on YouTube where kids were touched by humanoids and were kicked by humanoids. You saw that also, probably. And I think this is unacceptable.

11:57 Roberto D'Elia: So therefore, I would describe it as: innovation should happen above the safety floor, not instead of the safety floor. So this should be the perception, in my opinion.

12:07 Interviewer: The event brings together European and Chinese OEMs, integrators and buyers all in one room. Who are you most keen to meet in Munich?

12:16 Roberto D'Elia: Actually, I'm most excited about the conversation between the different groups in the room. You have different perspectives. Of course, extremely interesting is speaking with Chinese OEMs. China is currently moving extraordinarily quickly from development towards industrial and large-scale production. And I saw it on the ground in February. I mean, they can learn a lot about the speed of China's industrial ecosystem and willingness to deploy technology. I think the Europeans are lagging in that.

12:50 Roberto D'Elia: But I'm quite interested in speaking with European operators and integrators, because I think this part, to be an integrator, will be much more interesting in the future, because you are between the OEMs and the user. You will be in the middle of that.

13:06 Roberto D'Elia: For the OEMs, probably I will be asking also: how can we manufacture 10,000 robots? I mean, this is something that the OEMs are now faced on. They want to scale as much as possible. The operator will be asking: how can we deploy these robots in a safe and real-world environment?

13:28 Roberto D'Elia: And there we see the role of Humanoid Robotics Association, where we bring all these topics together in one place. And this is important to steer this discussion, to put it together, because they have their single purpose, every one of them, and we try to bring them on one table and then discuss these topics. And for me, Munich is therefore less about one meeting on a particular company, but much more creating exactly those conversations in between these different aspects. Because the next phase of humanoid robotics will require an ecosystem, not a better robot.