Latest / CXOTalk / Robots and Physical AI: Strategy and Technology 2026 | CXOTalk #905
Transcript
- 0:00An AI chatbot that hallucinates is annoying, but robots that
- 0:04screw up can cause death and injury.
- 0:08Burkhardt Bookham is CTO of Hexagon, which builds physical
- 0:13AI. Physical AI is actually AI that
- 0:17shows up in the real world. It is embedded in machines that
- 0:21move, operate and affect people, assets and operations.
- 0:27Physical AI let's autonomous systems like drones or robot
- 0:31perceive, understand reason and performs actions in the real
- 0:37world. The core aspect here is the
- 0:42operating in the real world. And when we talk about the AI
- 0:46aspect, where does that get overlaid into this?
- 0:50It's important for physical AI and, and specifically you just
- 0:54mentioned if a chat bot is hallucinating, it, it it's
- 0:58annoying. But if, if a humanoid for for
- 1:01example, is hallucinating or getting out of control, it's,
- 1:06it's, it's, it can be dangerous. So basically the bar is much
- 1:10higher for physical AI because it, it affects the real world.
- 1:15That means you have to ground your AI in in reality.
- 1:22That means you use calibrate sensors, verified maps, do
- 1:27accurate reality capture in order to create a exact replica
- 1:34of the real world into digital twin.
- 1:36And you constantly have to update this.
- 1:38So it's, it's Evergreen. And the system needs to be
- 1:42anchored in the truth so that a robot knows, perceives things
- 1:47right and can, can, can handle things accordingly or, or
- 1:52perform action and task accordingly, but into the real
- 1:57world. And therefore you need the
- 1:59strong connection into the reality or the strong link of
- 2:03the digital twin to the reality. You've used the term digital
- 2:07twin a couple of times. How does the digital twin aspect
- 2:14relate to physical AII? Would say it's the base because
- 2:20robots need to to to learn and they do it in various techniques
- 2:25and technologies, imitation learning, reinforcement learning
- 2:28and and so forth. But I, I would say the, the,
- 2:33the, the most straightforward and the, the ones that is, is is
- 2:40laying the bar highest for function safety.
- 2:43For for instance, is, is if you, if you take the reality in the
- 2:48real world and, and map it one to one into a digital reality
- 2:53with a with a precision and accuracy that, that guarantees
- 2:57you the ground truth that is is needed for the for the actions.
- 3:01And then ideally your robot hasn't on board a lot of sensors
- 3:06that first of all perceive the world and then basically can can
- 3:13also control the manipulating or the handling or the dexterity of
- 3:18the robot, all the locomotion where the where the robot is, is
- 3:22going. You have used the term ground
- 3:27truth. Yeah.
- 3:29What do you mean by that? We're really covering the the
- 3:33basics of physical AI and then of course we'll get more into
- 3:38the details, but this is super helpful to give us an
- 3:41understanding of as CTO of Hexagon, which develops these
- 3:46products to give us a basic context.
- 3:49What is important and also where you base afterwards your
- 3:54simulations on and is, is to have an understanding of the
- 3:59real world, of the, of the real environment.
- 4:01And you can do beautiful things in, in, in terms that if you if
- 4:07you accurately create a digital twin by reality capture
- 4:12technologies, by vision technologies for, for for, for
- 4:15example, and this is dimensionally accurate, this is
- 4:19accurate in scale and so forth. And then you can, can actually
- 4:23train the robot in this digital twin and, and try things out,
- 4:29make errors and then failures in the, in the digital twin.
- 4:33And then when you deploy it into the real world where you trained
- 4:38it on the digital twin, it's, it's, it's far more easy to, to,
- 4:44to do this. The robot directly knows the
- 4:49environment, perceives the right things.
- 4:51And what I also find it's it, it, it has to have a certain
- 4:57sensor stack on board because when you created the digital
- 5:02twin and we say for autonomous system, the digital twin should
- 5:06be autonomously updated, but that only works to a certain
- 5:11extent. So for the, for the things that
- 5:14changed, for example, a table or a chair that that have been
- 5:18moved or basically a workplace that looks slightly different at
- 5:22the time when you generated the digital twin, The robot needs to
- 5:26have enough perception sensors and reality capture sensors on
- 5:30board in order to perceive the difference and, and act
- 5:34accordingly. But I think the, the dimensional
- 5:38ground truth, and I think that's that's where the interaction
- 5:43between the matter, the workspaces, things you do and
- 5:48also collaborating with with with actual humans start is, is
- 5:54the ability to, to base it all off of, of, of reality.
- 6:00And I think that's also a hard problem that we need to solve.
- 6:07Furthermore in the future is to close this gap between the, the
- 6:12reality, the digital twin and then the digital twin is, is the
- 6:16sandbox where you simulate and train on in order to, to close
- 6:20this real to SIM gap again when deploying into into reality.
- 6:24Because robots in in essence then live in reality, in the
- 6:29real world. Folks, if you're watching on
- 6:33LinkedIn, pop your questions into the LinkedIn chat.
- 6:37If you're watching on Twitter X, use the #cxo talk.
- 6:42Burkhardt. Can you talk about some of the
- 6:45differences between AI, digital AI on the computer versus what
- 6:52happens when AI operates in the physical world in areas like
- 6:56safety, cost, reliability? It's also two very different,
- 7:03different things, because we all know the, the, the digital world
- 7:08and we, we, we, we know the, the agents and the, the Ted bolts
- 7:12and, and, and the copilots and so forth.
- 7:15But let's focus for a for a second on the, on the, on the,
- 7:20on the physical world where robots and physically I make it
- 7:25a difference or need to make a difference is is basically if
- 7:29they solve and and have humans to to collaborate and and do
- 7:36meaningful things. And, and here we, we talking
- 7:39about a very different metrics for, for, for example, we, we're
- 7:48talking about predictability, we talk about uptime, we talk about
- 7:54safety. And that's something that needs
- 7:59to be reliable modelled into, into the AI.
- 8:05And as I, as I said before, we, we're talking about maintain and
- 8:11security work flows that that has to have to be fulfilled in,
- 8:19in order. So, so, so, so it's, it's, it's
- 8:22not about showing the next coolest trick.
- 8:25Also, I like this very much. If I, if I, if I see a robot
- 8:29doing, doing something stunning or I have a very liquid motion
- 8:33and these kind of things. No, it's, it's, it's, it's
- 8:35probably first, first and foremost about the uptime, the
- 8:41the reliability and, and performing the workflows we, we
- 8:45intended it to, to, to do. So it's really all about the the
- 8:52business applications and being very clearly focused on on the
- 8:58business problems that are trying to be solved.
- 9:01Absolutely. You see, of course, a lot of
- 9:03YouTube videos where, where humanoids or robots are launched
- 9:08that do super cool things. And it's, it's so liquid these
- 9:12days. But I think the, the really hard
- 9:15tasks look actually a little little bit more boring to to, to
- 9:22be honest. Is is is working in, for
- 9:25example, material handling, checking things, quality
- 9:29inspection. What, what, for example, we have
- 9:33shown when we launched our, our humanoid was oh, humanoid
- 9:39scanning with a, with a laser scanner of the, the, the door of
- 9:45a of a car for for defects. It's probably far less cool and
- 9:52cool to watch like a robot is dancing or dog doing somosaurs
- 9:56and these kind of things. But, but these are the things
- 9:59that we expect specifically humanoids to, to do in the
- 10:05future to, to, to solve the tasks that are tedious, maybe,
- 10:10maybe dangerous, maybe in scenarios where you have a, a, a
- 10:15high degree of repeatability, but the workplace is changing
- 10:20and, and, and these kind of things and, and, and, and
- 10:24probably less about general purpose.
- 10:28It's, it's, it's doing jobs, but being very flexible in, in doing
- 10:34jobs. Otherwise, you could install a
- 10:38stationary robot for for just doing one or few things and
- 10:43probably will will outperform what we currently see in in
- 10:47humanoids. Anthony Scrifignano, who is a
- 10:51data scientist and a prominent data scientist and he has been a
- 10:55guest on CXO Talk a number of times, He actually asks about
- 11:00physical A exploits by malefactors and you mentioned
- 11:05security earlier. So can you talk a little bit
- 11:07about that aspect? It starts already with the
- 11:10design of the of the of the whole robot and, and it's also
- 11:15basically in, in, in in how you guarantee functional safety
- 11:21along the way in all the aspects in terms of it starts.
- 11:26Also, I, I lately looked at a, at a report where we have shown
- 11:30why, why are the, the current humanoids not giants and, and
- 11:34and these kind of things. So it so it started in building
- 11:39in things and building separating also also the the the
- 11:43control loop from the from the AAI, the physical AI parts of of
- 11:50of it. But you have to be in a in a
- 11:53position. And, and probably the best
- 11:56design goes in a in a direction is that that the robot does know
- 12:02when it does not know anymore, when it when it's lost and
- 12:06slowing down, stopping instead of of doing something, going on,
- 12:12tripling over these kind of things.
- 12:16And also it's important by by design that that that you design
- 12:22design quite fails. If you have that you have
- 12:25independency, redundancy in in what you, you do.
- 12:29And, and also have a have a sense of stack that allows you
- 12:34to, to perceive if things go basically out of the range where
- 12:40you, where you trained it for, whether it's be by reinforcement
- 12:44or imitation learning or other or other techniques in, in, in
- 12:48physically AI. But, but I think what is what is
- 12:53important is that you control all the stack.
- 12:56It comes from the locomotion, the dexterity, the, the sensor
- 13:01stack and then basically the physical AI that, that this is
- 13:05really fail safe because it could, could end up quite badly
- 13:10in a, in a disaster. If, if, if you triple over on
- 13:13some on, on, on something and, and, and people, people might
- 13:18get hurt. And that is basically to avoid
- 13:21it at all cost. In terms of of, I would say the
- 13:24big theme that is coming in 2026, which is functional
- 13:27safety. Unlike digital AI, where in most
- 13:33instances an exploit can have an impact on a a network or have
- 13:41financial consequences, with physical AI there can be
- 13:48physical damage that's caused. So you have this additional
- 13:52layer of complexity. Yes, absolutely.
- 13:56Manufacturers or, or development teams of, of, of specifically
- 14:01humanoids or quadruplats or or or drones or any autonomous
- 14:06system that, that, that that interacts and is in, in
- 14:10inherently dangerous in, in, in this regard.
- 14:15You have to, to, to be able to, to, to control the whole system,
- 14:20not just the model. So it's about the sensors, the
- 14:23compute, the controls integration, the maintenance
- 14:27even it's nice if it's working out-of-the-box, but but what is
- 14:33if, if there's wear and tear and overtime systems maybe not
- 14:37performed through the standards you, you had them at the
- 14:40beginning and then security. So all the work plus that these
- 14:45workflows determine whether it it, it actually works in the in
- 14:49the in the real world. So I would say the bar is far
- 14:53more higher than than in in, as you call it, digital AI.
- 14:58Greg Walters asks the question that I think lots of us have,
- 15:02which is at CES Computer Electronics Show.
- 15:08There have been lots of AI announcements from Atlas and
- 15:12Gemini to NVIDIA. Do you have any comments and
- 15:17thoughts about all the announcements that took place at
- 15:21CES? It's hard to comment on, on, on
- 15:24everyone on its on its own because there there's so many
- 15:28what, what, what I what we currently see.
- 15:31And when we, when we started for, for, for instance, our, our
- 15:35humanoid project, and it was basically based off of our
- 15:40experience we made with autonomous drones and, and what
- 15:46we, what we made of Kudrow plats, roducts as the robot
- 15:49dogs, as we was, we like to, to call them that probably.
- 15:54And, and these announcements show the way and probably with
- 15:58a, with an ever increasing pace that when we, when we, when we
- 16:04started this, we, we, we thought, how in the world are
- 16:08we, are we solving all this control loop tasks that, that we
- 16:13have to do? And, and with the, with, with
- 16:17all the announcement, all the technology that, that, that,
- 16:20that came it, it brought more and more tools to, to, to, to
- 16:27basically build, build robots. And I would say currently we
- 16:31are, we are, we don't see the end there.
- 16:33And I'm, I'm what, what we, we saw for NVIDIA the, the, the,
- 16:37the physical AI announcement around physical AI and as
- 16:43Nvidia's a good partner for, for us, we have been on the tractor
- 16:46trajectory with them for, for a longer time now.
- 16:51But, but it's, it's, it's, it's amazing to see what is the
- 16:57capability today of physical AI and, and, and humanoids.
- 17:03Yet I would say 26 and 27 is probably the years where we
- 17:10maybe step back a bit and, and, and, and, and look for deploying
- 17:17robots in actual manufacturing scenarios in, in, in, in, in
- 17:24workspace scenarios in order to, to, to, to, to solve real world
- 17:31problems. Because we have seen now that
- 17:33that locomotion is great. The liquidity we see in the
- 17:38motion of the fluidity of, of, of robots is, is stunning.
- 17:43It's so stunning that you probably need to take them apart
- 17:46on the stage that you see that, that there's not a human in a,
- 17:50in a, in a costume, but yet to be seen our task.
- 17:56I have a box full of gearbox of, of, of gears, for example, and
- 18:01there are, there are maybe maybe 20 in there that don't belong in
- 18:07the box. Can a, can a humanoid sort them
- 18:09out by by looking at them? Can they measure it?
- 18:13Can they see whether it's intolerance or yes, can it can
- 18:17be taken further? I, I think that's, that's yet to
- 18:21be proven. These, these use cases that
- 18:26these use cases can be fulfilled.
- 18:28I'm an absolute optimist in terms of robotics and, and
- 18:33humanoids. And I think it's absolutely
- 18:35possible. But I, I think, I think this
- 18:38year and the next year is probably the ones where we,
- 18:41where we see a lot of robots, humanoids specifically deployed
- 18:45in there's real world scenarios, maybe, maybe, maybe still
- 18:51isolated from the actual workforce.
- 18:56And then then doing this task. And you probably need then a
- 19:05more refined physical AI to, to do specific things, starting
- 19:12with, with, with various, various interactions already
- 19:16sorting and picking things. Is, is, is is one that that's we
- 19:22we should see that is that is quite doable in the in the in
- 19:26the future. Let's jump to another question,
- 19:31this time back to Twitter to X. You can see I love taking the
- 19:37audience questions like you guys in the audience are so smart.
- 19:42And by the way, now would be an excellent time to subscribe to
- 19:46the CXO Talk newsletter. Go to cxotalk dot dot com right
- 19:50now, subscribe to our newsletter, and we'll keep you
- 19:55updated because we have really incredible shows all the time
- 19:59coming up and you should join us.
- 20:01OK, here is a question from a technical question from Chris
- 20:05Peterson on Twitter, who says, do you see embodied AI and
- 20:12robotic systems using more centralized cloud, distributed
- 20:19edge or both? So is it cloud, edge computing
- 20:25or both? I would say both the cloud part
- 20:28is is as as as we said in the very beginning of the of the
- 20:32discussion is is very important when we when we talk very big
- 20:36data in order to to create digital twin and then the
- 20:41sandboxes for the the the robots to to to learn and then also in
- 20:47order to orchestrate maybe in the future.
- 20:50Currently we have a limited amount of of, of robots, but
- 20:55when there's a large amount of robot, you need also to to
- 20:58orchestra the, the fleets. And that's probably your your to
- 21:03do with an additional twin environment.
- 21:08However, to, to be fully autonomous to, to, to, to fulfil
- 21:12task and, and basically also learn on the job.
- 21:17I think the edge edge AI is, is decisive.
- 21:21That's why I think it's, it's also a success story to, to have
- 21:26sufficient compute in, in, in, in your, in your, in your robot.
- 21:33But I think together with something probably that that's,
- 21:39that is is in the direction of Federated learning, Federated
- 21:43AI, where one robot learns everything for the fleet.
- 21:48You you probably need to, to find a good balance between the
- 21:51cloud and the edge. When, when, when there's then a
- 21:57deployment of a, of a whole fleet.
- 21:59But I, I think in, in, in, in humanoid, specifically in or, or
- 22:04or robots in general, I think HAI is, is is really decisive
- 22:10for them because the more compute, the, the more things
- 22:14you, you do, it's, it's going more into the direction of a, of
- 22:19a, of a multi purpose robot, which I think is still a little
- 22:23bit out. But but I, I think hai is is is
- 22:28is is really important to have to have enough computer on the
- 22:31edge. This is from Jyotiram Ladd, who
- 22:36says most responsible AI regulations are written with
- 22:41digital AI in mind. Do you think today's regulatory
- 22:45frameworks are truly fit for physical AI, where AI decisions
- 22:50directly affect the real, the physical world?
- 22:54And Anthony Scrifignano asks the same question.
- 22:59Essentially, he says, are there emerging regulations beyond
- 23:02general AI in the context of physical AI?
- 23:06So both of them want to know about regulations and
- 23:11responsible AI and how is it different when it comes to the
- 23:14physical AI. What we see now is, is, and I
- 23:18recently I, I gave an interview about this function safety and
- 23:24the upcoming regulations. And I, I dig deeper into the,
- 23:28the topic, maybe, maybe to, to keep it, to keep it on the on,
- 23:33on, on this level. What we, what we currently see
- 23:36is, is probably what we've seen with the, with the UAVs and the
- 23:40drones a couple of years back. That why the technology is
- 23:44emerging and the capabilities and the skills are, are, are
- 23:48coming up. Then also the regulatory body
- 23:52will, will, will come, come further.
- 23:55We have, we have probably a very strict body already in the EU in
- 24:01order there, but but probably for for safety regulation in
- 24:06terms of physical AI. We see probably in this year
- 24:10and, and next year more, more regulatory bodies coming up or,
- 24:16or basically many, many manifesting.
- 24:21And I think that that function safety is, is, is still
- 24:26something that currently everybody's happy that the
- 24:30locomotion looks good and these kind of things, but needs to
- 24:34address that. And it's, it's it, it needs to
- 24:36address if you have a, a larger amount of, for example,
- 24:41humanoids working together, collaborating with, with humans.
- 24:45And I said, so in short, yes, I, I, I, I, I anticipate that, that
- 24:52we, we see a rise in, in the regulatory bodies for
- 24:56specifically physical AI because what we, what we have currently
- 25:01is, is machine guidelines, which probably are not that sufficient
- 25:05for the, for the autonomous systems, although they, they
- 25:08already cover fairpoint. And we have a a related question
- 25:13from Arsalan Khan on Twitter, who says AI algorithms use data
- 25:20to make decisions. In non physical AI, bias can
- 25:24potentially be corrected by updating the algorithm and the
- 25:28data. How do you tackle bias in
- 25:32physical AI? And he's asking also whether we
- 25:34need regulations. He's also interested in the
- 25:37regular regulation question. So the issue of bias in physical
- 25:43AI. It starts already with the
- 25:46accurate representation of the of the of the real world in in a
- 25:51digital twin. If you if you basically solely
- 25:56base your deployments and your trainings on synthetic data
- 26:01without dimensional accuracy and without the the right scale
- 26:05factors and so forth. I think that this is where you
- 26:09introduce the bias. If you, if you train in
- 26:14scenarios that are are based on on digital twins, which are
- 26:19dimensionally accurate and and and a an exact replica of the of
- 26:29the physical world, I think you can limit this, this biased.
- 26:33And then it's, it's, it's also one thing that the, the robot is
- 26:40itself is capable of, of sensing the bias in, in sensing whether,
- 26:49yeah, for example, whether the the perceived step height is,
- 26:54is, is XYZ with enough sensing power on on board.
- 26:59And that, that also influenced the, the, the compute.
- 27:03Then I, I think foremost these, these are the other, other
- 27:08things. And then basically, once you,
- 27:10once you've found a bias, you have to close the loop and, and,
- 27:14and, and bring it back. But that's something to be
- 27:17honest that we did in other industries and in other
- 27:20technologies since quite some time talking about, for example,
- 27:25the quality inspection of a of a, of an, of a car for, for, for
- 27:30example, there basically you, you close the, the loop from
- 27:34from the cut drawing to the machine part, you inspect.
- 27:38And if it's out of tolerance, you, you, you alter your machine
- 27:41to it. And I think maybe this is very
- 27:45very mechanical explanation of this but but it's same and needs
- 27:49to hold true here as well. Let's go to another question.
- 27:53This is from Sai Panamuru. And Tsai says how close are we
- 28:00really to autonomous operations? And what is the one thing still
- 28:06preventing us from getting there?
- 28:09And Greg Walters asks a very simple, very related question.
- 28:16He says this is the next stage. Embodied AI is general once it
- 28:21has all the six senses of a human.
- 28:24How close are we to this? So basically, how how close to
- 28:30autonomy and to this idea of robots as really helper
- 28:38humanoids? Very interesting topic and a
- 28:42good and good questions in in this.
- 28:44And I think that's what we, what we currently see is probably
- 28:49still still of the humanoids and and demos and so forth are, are
- 28:55tailor operated in, in, in, in, in these kind of things.
- 29:00However, I think, I think probably we are in certain tasks
- 29:04not that far away from, from, from autonomy if we basically
- 29:09constrained the environment and to, to, to some extent for for
- 29:15me, it depends a little bit on, on what is the capability of the
- 29:22sensor stack of the of, of the robot.
- 29:27That said, we, we, we have currently also also a lot of of,
- 29:36of imaging devices, lot of time of flight cameras, microphones
- 29:42and so, so forth in the, in the, in the sensor stack.
- 29:47The sensor stack is not yet fully, fully there where where
- 29:52basically humans, humans are to my knowledge in, in, in, in what
- 29:57we what we currently see on the on the market on the prototype
- 30:03stages. But I would say strong vision
- 30:07systems, strong lighter systems combined with, with some
- 30:14speciality systems that, that do for example, touch sensitivity
- 30:20in the, in the end, effectors enable us already now to, to, to
- 30:26basically autonomously perform certain task, for example,
- 30:33picking things and, and, and, and and placing things.
- 30:38I would say there we are already quite advanced probably in the
- 30:44in, in the longer term, because I, I think the whole humanoid
- 30:50businesses was, was quite fascinated with, with locomotion
- 30:54and, and basic dexterity. Right now I think we see in the
- 30:59future much more census being built in and that that enables
- 31:02then autonomy. Probably drones are a little
- 31:07further advanced what we what we what we've seen here in in this,
- 31:13but also the interaction with Mattis is of course less here,
- 31:19but but I would carefully estimate there that we are,
- 31:24yeah, maybe, maybe 1-2 Max three years away for autonomy in
- 31:29certain constrained areas. I would be very careful in
- 31:34saying general purpose everywhere or robot in a
- 31:37household can do everything. I would be careful with this.
- 31:40But in in in constrained environments it's it's
- 31:44foreseeable. Now, can you talk about the
- 31:49timeline for robots to create actual economic value, not just
- 31:56demos? And of course, that's relating
- 31:58to related to what you were just discussing.
- 32:00There are various stages and phases where, where, where,
- 32:04where we're in. Probably what we've seen in the
- 32:08in, in the last couple of months and maybe maybe year is a lot of
- 32:14announcements with robot manufacturers and I would say
- 32:21early adopter companies, whether it's it's automotive and and so
- 32:27forth. And and, and, and I think in
- 32:29this stage, certain use cases were tested in a lab
- 32:35environment. And I would I would I would see
- 32:39currently that that this goes out now in onto a true
- 32:43manufacturing side, but not in the in the quantities of dozens
- 32:49of of robot maybe, maybe a small amount in a, in a, in a, in a,
- 32:54in a constraint part of the of the factories.
- 32:57I think that that what we can expect fully this year, Yeah,
- 33:01in, in, in this and then a more a more rolled out deployment
- 33:09probably probably than than than the year after, but still to
- 33:16some extent constraint and to some extent limited to to a
- 33:24certain certain use cases. Can you describe where robotics
- 33:31deployments today most often fail?
- 33:35Is IT technology limitations? Integrations with existing
- 33:40systems? Organizational readiness?
- 33:44Where? Where are the obstacles today?
- 33:46Probably the technology is, is is already quite advanced in in
- 33:53in this, but it's it's, it does not have a track record yet in
- 33:57terms of deploying hundreds and thousands.
- 34:03So basically all the all the teething errors, the teething
- 34:07problems that you that you would get with a with a bigger, bigger
- 34:11fleet and all the maintenance and and all the the
- 34:14serviceability topics have probably not addressed in in in
- 34:20in something. If you compare it for, for
- 34:22examples, to to to to drones or even in higher scale cars at, at
- 34:31this at this point point in time, they're basically we, we
- 34:38we yet need to see how how basically how maintenance
- 34:43intensive such humanoids is. But probably the, the, the
- 34:48failure level is higher, not on the technology and the
- 34:51capability part, but but probably for I would see
- 34:56Enterprise is not yet ready for, for basically welcoming A
- 35:02humanoid fleet and welcome in it to, to work alongside with the,
- 35:09with the, with the work flows. We just talked about the
- 35:12regulations. I I think there we need to do a
- 35:15couple of steps more in order to have this this correct.
- 35:21And then basically on the on the, on the other side is is for
- 35:26me the the questions still open then how does enterprises
- 35:32corporates basically incorporate them the the humanoids alongside
- 35:38with with with workers in, in, in operators in an, in an actual
- 35:46factory or manufacturing discrete manufacturing
- 35:48environment? Probably, probably the latter
- 35:52part is the is the more decisive going forward because technology
- 35:56will evolve and be we would see it quite quickly.
- 35:59And Greg Walters on LinkedIn asks a question directly
- 36:03relating to this. He said aren't humanoid robots
- 36:08in human like form soon to be obsolete in the manufacturing
- 36:13realm as new production lines will be designed to be run
- 36:18without humans? We have a division that calls is
- 36:22called manufacturing intelligence and they're working
- 36:24towards what is called lights of manufacturing, where basically
- 36:30manufacturing is or the manufacturing environment is
- 36:34designed for having no humans at all in the in the loop and and
- 36:40even even even taking the lights off in order to it runs
- 36:44autonomously the the whole the whole manufacture.
- 36:49I think for these environments, you, you probably would not
- 36:54develop a humanoid for. However, there's a still a world
- 36:59that is made for humans. There's still a large legacy all
- 37:04over, all over the world where basically sometimes whole shifts
- 37:11breakdown because the machine to adjust to have to, to be
- 37:16resetted and, and these environments are are made for,
- 37:23for for the last decades for, for humans.
- 37:26So I think the proportion, the height and everything is, is
- 37:30still very, very beneficial for a human.
- 37:33But I give you 100% right, If you, if you design a fully
- 37:37autonomous lights off manufacturing plant, then
- 37:43probably humanoid would be not the desired form factor.
- 37:48But for everything else and there's a large legacy still
- 37:51existing. I would say the humanoid is a,
- 37:54is a, is a very, very nice form factor for this.
- 37:57But because we have seen it in the, there was an early hide
- 38:01for, for this robo docs and and and and we developed scanners
- 38:07that went on on, on on robodocs. It's sometimes did not work
- 38:13because there was an elevator button to to push or it was not
- 38:21the right point of view for for the system.
- 38:25These are the the human arts, of course, over over overcoming.
- 38:29But I give you right for for a autonomously designed
- 38:35manufacturing place, you would probably work differently.
- 38:38So the the form, the humanoid form, could we say that it's
- 38:45essentially the user interface? I would say, but well, basically
- 38:51it's, it's the the embodiment that resembles the user
- 38:56interface that we are used to in, in, in, in this regard,
- 39:01yeah. So we have the robotic system
- 39:05that is independent of the particular form in which it's
- 39:10embodied, right? You could have a robotic system
- 39:12that's inside a drone, something flying or a dog or, or humanoid
- 39:19or, or a machine in a factory, essentially, right?
- 39:25So it's the packaging, it's the external packaging.
- 39:28Absolutely. And I, I, I think what we, what
- 39:31we still see is, is that the in, in, in the humanoid there,
- 39:37they're not trying to, to, to, to exaggerate certain qualities
- 39:43of a, of a, of a human not making it, it, it much faster,
- 39:46not not longer legs, not slightly longer arms, these kind
- 39:50of things. Now we see a quite quite an
- 39:53quite a nice representation of a human as far as you want to go
- 40:01with without going into the uncanny valley situation.
- 40:06Talk more about digital twins. You described earlier how the
- 40:11digital twin concept is foundational, but how are
- 40:17robotic and humanoid digital twins different or more complex
- 40:21from those used in industry, for example, in jet engine
- 40:25maintenance operations, which is an example that we hear about
- 40:28all the time? What is is is very important for
- 40:32these autonomous systems and the and the and the digital twin
- 40:35and, and for example, certain digital twin is always fit for
- 40:41purpose. For example, if you, if you want
- 40:45a digital twin of a jet engine, then then then certain qualities
- 40:48need need to be really, really in this digital twin in order to
- 40:52drive a simulation, in order to manufacture it, in order to lay
- 40:57out your production of the blades and blisk and whatever
- 40:59you, you, you, you need for this.
- 41:02What is important for, for for the digital twin that you have
- 41:09in, in, in, in the, in the space of of of of robotics is that the
- 41:15autonomy of these systems requires that these digital
- 41:19twins have to be alive. Of course they need to have the
- 41:22geometry, physics and, and, and the, the right scale.
- 41:29However, you, you have to acknowledge the sense of
- 41:34behaviour, they, the dynamic obstacles that there are.
- 41:42And, and basically they have to be constantly updated because,
- 41:48yeah, digital twin and yeah, famously, I think the, the, the
- 41:53probably the most famous digital twin years back was, was
- 41:58basically the digital twin that allow the, the, the
- 42:01reconstruction of Notre Dame in, in, in, in Paris.
- 42:06That was basically by accident or by luck, we had a digital
- 42:09twin of, of, of, of this, this is an absolute static one.
- 42:14And you want to rebuild a building of this digital twins
- 42:19for for robots needs to be Evergreen, always updated and
- 42:24and sometimes even in in, in in real time.
- 42:28And that's what what different stem from, for example, the
- 42:32digital transfer for a part that you want to manufacture.
- 42:37So Anthony Scrofignano is asking about collaboration thinking
- 42:44among the robot. So he's thinking, for example,
- 42:50drone swarms. Can you talk about that aspect
- 42:54of it? That's something that is
- 42:56essential working in fleets, collaborating robots to robots,
- 43:01drones, drone system, like like in, in, in, in swarms as, as I
- 43:08said, coming back to the, to the, for example, autonomous
- 43:12drones when there's a communication or basically
- 43:16autonomous vehicle in mines. Yeah, the, the communication
- 43:22between the robots, the autonomous systems is, is
- 43:25essential and it's essential for fulfilling the task.
- 43:29Sometimes it's also essential for for learning together in
- 43:34terms of Federated learning. If if one robot is is is is is
- 43:40training something new, then basically you want the the whole
- 43:43fleet to to to have the skills as well.
- 43:47Let's talk about some of the business implications here.
- 43:49What do boards of directors typically misunderstand about
- 43:55investments in robotics and physical AI?
- 43:59But I think many, many boards overestimate the speed and under
- 44:05mask the system work. It's, it's, it's, it's not a, a
- 44:10software roll out. It's, it's a complex engineering
- 44:14system. That's what also we we saw from
- 44:17when we talked about the regulation from digital AI to
- 44:20physical AI. It's safety engineering, it's
- 44:23operation change, it's integration and service and
- 44:26maintenance. And you have to think the whole
- 44:30the whole cycle through in in in order to to to execute on on a
- 44:37robot investment. If you were advising a board
- 44:42considering a major robotics investment, what is the one
- 44:45thing you would insist that they understand before approving it?
- 44:49I would insist on clarity about the job to be done and the risk
- 44:53envelope. I think you clearly need to know
- 44:57the task, the environment, safety constraints and what
- 45:00happens when confidence drops. And I would also insist they
- 45:05treat data and digital twin readiness as a prerequisite.
- 45:09If you do not have a a reliable picture of the reality you're
- 45:13in, you're probably funding A perpetual pilot.
- 45:18Robotic efficiency or robot efficiency can mean job losses
- 45:24and displacements. Can you talk about the future of
- 45:30work in a increasingly robotic and humanoid robot world?
- 45:38In an idle scenario where we where we currently work towards
- 45:42through is the is the collaboration between humans and
- 45:46and and and robots. And we would like the humanoids
- 45:51to do things that are quite mundane, that are tedious, that
- 45:56are, are dangerous, that are risky going into situations,
- 46:01whether it's mining or, or in the nuclear where humans should
- 46:06not go. So basically the, the working
- 46:09together. And in, in this scenario, I
- 46:13would say we, we, we, we upskilled the humans to a, to a
- 46:17different level. And I just read an article about
- 46:21a famous roboticist and said, I'm, I'm optimistic about
- 46:24robots, but they should always work together with humans.
- 46:29And I believe in this as well. What does responsible deployment
- 46:35of robots therefore look like? For me it means safety by the
- 46:39design. You need the validation, the
- 46:42traceability, the fate safe. You need to have the the audits
- 46:48done, clear handover rules and and, and this is especially
- 46:53crucial in what I said before in the especially in the mixed
- 46:57human robot environments. And it also means for me human
- 47:01centred integration, a training workflow redesign and the
- 47:06accountability to determine whether robots create value or
- 47:11or or maybe chaos in in in in here.
- 47:15Arsalan Khan asks the question, Do you see a time where robots
- 47:24become physical, AI becomes more and more autonomous, and will we
- 47:29need a manager? A manager AI that checks if
- 47:34worker AIS are supposed to be doing what they are tasked to
- 47:39do. Probably at some levels, yes,
- 47:42but but I would see that ultimately you should have
- 47:46somewhere a human in the loop that is, that is that is
- 47:51unavoidable. Whether it's then a management
- 47:54AI or basically it's a it's a management AI copilot that that
- 48:00helps human to to to watch over a fleet yet to be seen.
- 48:07But but, but I think the the, the, the the the layering stacks
- 48:12of of AI probably probably makes a lot of sense.
- 48:16Where do you see all of this going over the next two or three
- 48:22years? That enterprise deploy more and
- 48:25more autonomy in in constraint domains inspection into a
- 48:30logistics, picking, placing, handling, site monitoring and
- 48:35assisted manipulation with scale in brackets.
- 48:41When the environments are met mapped or you have digital twins
- 48:45and and the success matrix are are are really, really clear.
- 48:50What I would say it's a little bit far away and probably still
- 48:55out of reach is general purpose solving everything, complete
- 49:00human level dexterity, reliable operation in in, in, in chaotic
- 49:07spaces and so forth. Because probably even if the
- 49:14technology would be there, it it would probably be hard to to, to
- 49:19get the safety level and the safety certification right in,
- 49:23in this environment. So probably the the latter 1 is
- 49:27the is the hardest engineering problem.
- 49:30So I feel personally, the so-called Butler humanoid is
- 49:34probably still a way away. Are there one or two core
- 49:40technical challenges that you are very much interested in and
- 49:47focused on right now? If you have an accurate digital
- 49:51twin in order to, to do the training right, in order to
- 49:54close the simulation to reality gap as as narrow as possible,
- 49:59That's, that's, that's one of those Then basically the, the,
- 50:07the, the clarity on the use case and, and the clarity on, on what
- 50:13is out of scope currently. It's, it's, it's, it's one other
- 50:18thing. And then, then for me, it's,
- 50:20it's, it's, it's also what are the right sensors that I, I put
- 50:25on a, on a humanoid in order to fulfill all this task.
- 50:29And yet to have it as a as a cost level that is is creating a
- 50:35economic value and a sufficient ROI in order to to to to see the
- 50:42invest go through. Well, this has been such a
- 50:46fascinating conversation. Burkhardt Bocham is CTO of
- 50:51Hexagon. Burkhardt, thank you so much for
- 50:55being with us. It's been a really interesting
- 50:57conversation. Thank you so much, Michael, for
- 51:00having me. Thank you.
- 51:01And everybody, thank you for watching.
- 51:03Now before you go, now is the time to go to cxotalk.com and
- 51:09subscribe to our newsletter so we can keep you up to date.
- 51:13And soon we're going to be starting to send out summaries
- 51:16of our episodes on our newsletter.
- 51:19So check it out. We have awesome shows coming up.
- 51:24And everybody, we'll see you again next time.
- 51:27You guys, you guys are so smart. I always say that because it's
- 51:30really, really true. Take care everyone.