Best of LinkedIn: Robotics CW 38/ 39

00:00:00: Provided by Thomas Hallgeier and Frenis, based on the most relevant LinkedIn posts about robotics in calendar weeks thirty-eight and thirty nine.

00:00:08: Frenus is a B to B market research company that supports robotics and smart manufacturing providers with building feature by feature competitive intelligence That shows exactly how their product stacks up against the competition.

00:00:20: You can find more info in the description.

00:00:22: So welcome back to The Deep Dive everyone.

00:00:25: Yeah thanks for having me again Always good to be here.

00:00:27: Today we are really just breaking down the top robotics trends that have been blowing up across LinkedIn recently.

00:00:33: Right and specifically tailored for you, The Smart Build & Manufacturing Professionals out there.

00:00:38: Exactly I mean We're strictly focusing on the shift from those flashy lab demos to real world stuff.

00:00:45: Yeah moving way past the robots doing backflips for likes On YouTube

00:00:50: Because For manufacturing professionals the conversation has entirely shifted.

00:00:54: It's all about payload Uptime And actual factory floor integration now.

00:00:59: It really is the era of the movie set robot demo, it's basically over.

00:01:04: yeah and I want to start by looking at this humanoid reality check Basically function over form.

00:01:10: Gershon Selnicker had its fascinating post recently.

00:01:14: Oh

00:01:14: The Boston Dynamics comparison right?

00:01:16: Yeah exactly he compared their Atlas Robot from twenty nineteen To the twenty-twenty six version bulky hydraulic machine.

00:01:27: It looked like a construction vehicle,

00:01:28: it totally did just fighting high pressure fluid Just to do basic gymnastics right with all these wires hanging out

00:01:35: Right.

00:01:35: but then you look at the twenty-twenty six version and it is All electric.

00:01:38: yeah its superfluid And has zero external wiring?

00:01:42: Its Like A completely different paradigm.

00:01:44: it Is in.

00:01:45: Ronald Van Loon and Dr Marcel Volmer both jumped In To add some context to this Because Atlas now handles thirty kilogram loads repeatedly.

00:01:54: Thirty kilograms, wow!

00:01:56: Yeah over and over again because the new benchmark isn't you know.

00:01:59: look what it can do.

00:02:01: It's What Can It Do Every Single Day In An Existing Messy Factory Workflow

00:02:06: Which Is Really Changing The Literal Shape Of The Robots We're Building?

00:02:09: What You Mean?

00:02:10: Well Aaron Prather Pointed This Out.

00:02:11: He Noted That Companies Like Aptronic Toyota And Even Agility Are Actively Embracing Wheels Over Legs.

00:02:17: Now

00:02:18: Oh Interesting Wheels instead of walking.

00:02:21: Yeah, because the race is no longer about building a robot that looks exactly like us.

00:02:25: You know it's about building up body That fits this specific industrial

00:02:28: work.

00:02:30: But let me push back on that for a second.

00:02:32: sure if form follows function right and wheels are just vastly more efficient For a flat factory floor Are we maybe over indexing on humanoids entirely?

00:02:42: That Is a very valid question but you have to look at the sheer scale of humanoid manufacturing that is already underway.

00:02:49: It's kind of answering that question for us.

00:02:51: Oh, you mean like the UB tech numbers?

00:02:53: Exactly!

00:02:54: Peter Kapp has highlighted this.

00:02:55: UB Tech's Liuzhou factory in China is currently producing one industrial humanoid every ten minutes.

00:03:01: Every Ten Minutes?

00:03:03: Every...ten minutes.

00:03:04: yeah That's ten thousand humanoids a year from ONE facility.

00:03:08: Wow

00:03:08: And then Robert Little noted Toyota's estimated push.

00:03:11: They're looking to deploy four hundred thousand robots, spending roughly six point four billion dollars a year starting in twenty-twenty eight.

00:03:19: Six point for billion?

00:03:20: I mean

00:03:21: wrap your

00:03:22: head around those numbers but you know Tim Iscock actually brought in really sobering reality check on these massive deployments.

00:03:29: Oh

00:03:29: right the agility digit five launch.

00:03:31: Yes

00:03:32: So Agility just launched with over three hundred million dollars in orders which is great.

00:03:36: And they disclosed the payload, battery life and charge ratios all normal specs.

00:03:42: But they critically omitted something

00:03:44: Right.

00:03:45: They omitted all of safety data No force limits no stop times no incident data from their last three years of deployments

00:03:52: Which was wild

00:03:54: For a machine that's explicitly billed as being cooperatively safe.

00:03:58: That emission Just a massive red flag for any floor manager.

00:04:02: It absolutely is because you can't just slap an AI on the robot and call it safe, which actually transitions us perfectly into second theme.

00:04:09: we saw in LinkedIn...

00:04:11: The brains behind the brawn right?

00:04:12: Exactly!

00:04:13: The architecture of physical AI.

00:04:15: Because hardware really only has half the equation here.

00:04:18: What's making these robots intelligent enough to deploy its scale?

00:04:22: Well Nicholas Sauvage shared great insight about this.

00:04:25: He said that real moat in robotics isn't the hardware design anymore.

00:04:29: It's not.

00:04:30: No, it is the deployment learning loop.

00:04:32: A system only gets better when field evidence from real-world messy use cases is converted into improved models.

00:04:38: Oh right because hardware is basically just a commodity at this point.

00:04:42: Exactly And Antipatrix expanded on that with really crucial engineering detail.

00:04:47: You can't run standard vision language action model or VLA At ten hertz

00:04:52: Because its too slow for physical world

00:04:54: Way to slow Contact dynamics change in two milliseconds.

00:04:57: Two milliseconds Yeah.

00:04:59: So if a robot hits a rigid surface, A one hundred millisecond delay from a ten hertz model means the physical hardware takes the entire shock.

00:05:06: Which is how you completely destroy a million dollar actuator?

00:05:09: Right!

00:05:10: so The solution as Anto explains is layering.

00:05:13: You have your high level AI intent but layer it with five-hundred Hertz deterministic reflex control loops running directly on the Hardware itself.

00:05:21: That makes total sense.

00:05:22: It's like giving the Robot a spinal Reflex right Like pulling your hand off of hot stove.

00:05:26: Exactly

00:05:27: that.

00:05:28: And Tushar Makhar actually connected this to a real-world construction example that I thought was brilliant.

00:05:33: He's building a drywall spraying robot.

00:05:35: Oh,

00:05:36: talk about a messy environment.

00:05:37: Right!

00:05:38: A drywall seam is basically the width of thumb but wheeled robots base never parks in exact same spot.

00:05:44: twice on dirty construction site There's

00:05:46: dust everywhere.

00:05:47: it shifts

00:05:48: Yet constantly.

00:05:49: so they have synchronize at thirty hertz camera with five hundred hertz controller on this chaotic site just to hit that tiny seam.

00:05:58: That is an incredible engineering feat!

00:06:01: It really is, and hey by the way if you want to stay ahead of these kinds of technical shifts in manufacturing make sure you hit subscribe so you can catch our future deep dives.

00:06:10: Definitely do that because the software economics of this intelligence aren't just fascinating right now.

00:06:15: Jayman Shaw highlighted a company called Skilled AI.

00:06:19: Oh I saw his post.

00:06:20: The numbers were crazy.

00:06:22: Yeah, they

00:06:22: hit a one hundred million dollar recurring revenue run rate just ten months after their first commercial deployment.

00:06:29: Ten months to one-hundred million in ARR for hardware?

00:06:33: Well that's the thing.

00:06:34: because there model can observe and acquire physical skills The intelligence becomes a reusable layer across different hardware.

00:06:41: Ah okay

00:06:43: It finally brings those high software like margins into physical work.

00:06:47: You know it's so much like the early days of PCs where everyone is just sitting around waiting for an operating system to standardize all the messy hardware.

00:06:55: That's a perfect analogy!

00:06:57: And Ribbon Matthew pointed out that NVIDIA is actively trying be that OS.

00:07:01: Really?

00:07:03: Yeah,

00:07:03: they aren't building a robot... They are building shared infrastructure.

00:07:07: things like G-R-O-O T for learning and Isaac Sim for simulation.

00:07:11: companies like Figure Agility and KUKA are using it.

00:07:15: That is so smart.

00:07:17: But to prove these brains actually work, we need real tests right?

00:07:20: Not just those polished marketing videos We were talking about earlier.

00:07:23: absolutely and Nicholas Keller Just launched something called reality check To solve this.

00:07:28: what does that?

00:07:28: it's a public manipulation benchmark.

00:07:30: they did fourteen thousand four hundred Real-world rollouts to basically stop the industry from relying on cherry pick demos.

00:07:37: oh

00:07:37: wow Fourteen thousand four Hundred rollouts that Is exactly What The Industry Needs

00:07:42: It really is.

00:07:43: So If ninety-nine percent reliability is the gold standard for a factory floor benchmark like that, what happens when you move to an environment where even ninety nine point nine percent isn't good enough?

00:07:55: The stakes go way up.

00:07:57: Yeah let's transition to this final theme high stakes robotics specifically the economics and limits of autonomy in surgery.

00:08:05: Right, because D'Ella Ukhana shared that classic demonstration of surgical precision.

00:08:10: You know the robots peeling delicate kiwis or grapes?

00:08:13: Oh yeah those videos always go viral

00:08:15: They do.

00:08:15: it really shows off the tremor filtering and the micro movements.

00:08:18: It looks like magic.

00:08:20: but Dr Michael Mendegani completely pivoted the mood on this with some hard data.

00:08:25: What

00:08:25: did he find?

00:08:26: He analyzed three point-three million robotic surgical procedures.

00:08:30: Okay,

00:08:30: huge sample size!

00:08:32: Huge.

00:08:32: and out of those there is a device or instrument malfunction rate at roughly one point zero percent.

00:08:38: Wow I mean one percent.

00:08:40: in software Is a rounding error but in surgery?

00:08:42: Right when the patient is open on the operating table that One percent is absolutely terrifying.

00:08:48: That's thirty thousand malfunctions...that

00:08:50: is a scary number

00:08:51: yeah.

00:08:52: so Menegoni insists that humans must always, ALWAYS retain the abort decision.

00:08:58: Which Amadou Uiwara also built on?

00:09:00: He noted that in healthcare... ...the most important feature of an autonomous robot isn't actually what it can do!

00:09:06: What is it?

00:09:06: It's knowing when it is outside its limits and then handing control safely back to the surgeon.

00:09:11: Oh..that makes so much sense.

00:09:13: Yeah we usually measure AI by how much work we can offload.

00:09:16: right But in medicine The metric is how safely we take control back.

00:09:20: That is such a fascinating way to frame it But I want you to look at this from a business operations perspective for a second.

00:09:26: Okay, because it's not just about safety right?

00:09:29: It's about unit economics.

00:09:31: Oh entirely and Tiger Buford proved This perfectly.

00:09:35: he posted About high volume surgeon owned ambulatory surgery centers or ASCs.

00:09:39: And what are they doing?

00:09:40: They Are actively kicking one million dollar robotic arms out of the OR.

00:09:45: wait there getting rid Of them.

00:09:46: why?

00:09:47: Because The robot adds fifteen To twenty five minutes of setup and turnover time per case.

00:09:51: Oh, wow.

00:09:53: Yeah and outpatient surgery those lost minutes just completely destroy profit margins.

00:09:57: So it's forcing centers back to manual or hybrid workflows

00:10:00: Just because of the setup time.

00:10:02: that is incredible.

00:10:03: And you know speaking of destroying profit margins Lisa Verunkova added The ultimate cautionary tale Of robotics capital To this discussion.

00:10:11: Are you talking about vicarious surgical?

00:10:13: Yes

00:10:13: Vicarious Surgical.

00:10:14: they recently dissolved after twelve years of development.

00:10:17: Twelve Years Twelve Years.

00:10:19: They raised four hundred and twenty five million dollars, achieved a one point.

00:10:23: One billion dollar valuation had FDA breakthrough designation and even have Bill Gates as an investor.

00:10:29: And

00:10:30: they still went under?

00:10:32: Because the never reached design freeze.

00:10:35: The deadliest valley in hardware

00:10:37: exactly every single dollar when it to building another slightly better prototype rather than a manufacturable product

00:10:44: that just couldn't stop tinkering

00:10:46: right.

00:10:46: It just proves that the distance between a working demo and a frozen design is the hardest part of this entire industry.

00:10:52: Yeah,

00:10:52: it really does.

00:10:54: Well if you enjoyed this episode new episodes drop every two weeks.

00:10:57: Also check out our other editions on smart manufacturing in industrial AI digital construction and connected tools and equipment

00:11:04: And thank you so much for joining us and listening today.

00:11:07: We really appreciate you spending your time with us on this

00:11:09: absolutely.

00:11:10: and before we go I want to leave you with one final kind provocative thought to mull over based on all these posts.

00:11:18: Lay it honest.

00:11:19: so we've talked heavily about the AI models today, The brains of these robots.

00:11:23: but Erdem Uren and Leo Sue both point out that the real industrial base sits deep in the physical supply chain

00:11:30: right?

00:11:30: The actual parts?

00:11:32: yeah the gearboxes the actuators the sensors.

00:11:35: These take decades to build imperfect And currently they are heavily dominated by legacy manufacturers In Europe & China.

00:11:42: Okay, I see where you're going with this.

00:11:44: Yeah so if the software layer of AI is truly borderless You can download it anywhere.

00:11:49: Does the geography of hardware supply chain ultimately decide who wins global robotics race?

00:11:55: That's a massive geopolitical question to think about Definitely something to keep an eye on.

00:12:00: And hey don't forget to subscribe So that we might tackle our next deep dive.

00:12:05: For sure Thanks for tuning in and we will catch ya.