Inside a Humanoid Robot Company

Indexado: 2026-05-15

[00:00:00]We've seen the choreographed demos and cherry-picked clips of humanoid robots doing backflips and kung fu, but those don't tell us how the robot would handle an un-rehearsed role in a messy warehouse or a shifting production line.

[00:00:14]To find out what's actually ready for the real world now, I've come to Kinesyx in Bristol here in the UK, where they've developed the KL 1, a humanoid robot that foregoes the conventional method of walking to focus more on speed, precision, and onboard intelligence.

[00:00:30]While there is ongoing development here, this isn't a prototype for 2030. This is a practical solution for today.

[00:00:37]We will be taking a closer look at the KL 1 and the team that developed him, but who better to properly introduce Kinesyx and the KL 1 than the man that started it all, the CEO and co-founder Brennan Pierce. So, I've been obsessed with robots all my life. I've been doing it now professionally for 20 years. I was lucky to be in Bristol and they opened the first robotics lab, Bristol Robotics Lab. Um, what did I miss there?

[00:01:01]There wasn't really much of a robot industry, so then I spent a long time in academia designing humanoid robots around the world at like America, Munich. And then a couple of years ago started doing uh robotics as a startup world, primarily in like AMR for restaurants, and now obviously do Kinesyx for humanoid robotics. I started Kinesyx in Silicon Valley. There there's a pros and cons. One is there's all of this uh talent there, but B, it's very expensive, whereas England has such great talented engineers and not many companies, so it's quite easy to attract world-class talent, whereas in the Bay Area you have to compete with Facebook, Nvidia, and all these people that pay a lot of money. So, on that topic, you have a fantastic team of engineers and specialists around you here. Unless you're Tony Stark, people don't expect you to produce some mind-blowing robotics in a cave with a box of scraps, but how hands-on with the background you have with robotics, how hands-on have you been with the KR1 and the development here? Yes, so the V1 I made all on my own. So, I designed it all in CAD, did the 3D printing, did all the electronics, and also coded it.

[00:02:09]So, I got the first version using an AI model. So, you could actually speak to the robot, get it to pick up a can of Coke, and pass it to you. It took me about a year to do that on my own.

[00:02:21]Renn certainly has an impressive background in robotics, but it takes a village, and he's surrounded himself with an impressive team of engineers and specialists to work with him at Kinisi.

[00:02:32]So, Paris, um you're a mechatronics engineer, I believe. Is that correct? Yeah. What is it you do here? What are you really working on? Um at the moment, I'm working on this new gripper.

[00:02:44]Um it's picking up kind of much bigger, heavier items for some potential customers that we've got.

[00:02:49]Um and so, this is this is a kind of quick prototype that I designed and printed last night. So, I'm presuming you're using 3D printers heavily for R&D, but it allows you to work on new designs pretty quickly. Yeah, it allows us to iterate so much faster. Yeah, the 3D printing that we do in-house in the kind of bamboos, but then also we actually 3D print out of metal quite a lot of parts on our robot. When it comes to designing a new part like this, which obviously it's in like the prototype phase at the moment, what is your sort of priorities when you're starting out?

[00:03:16]Do you design the parts, focus on the functionality, and don't worry too much about like the size and it being too power-hungry, and then and then refine that in once you've got it working, or do you initially design it to work within the limitations that you know you're going to have to, and then improve the functionality to fit within that? Yeah, we kind of design to kind of try and address the kind of big risks early is is the important thing. And and depending on what we're designing, that kind of varies. So, for example, with this gripper um or grippers in general, most of the risk is around like functionality. Like, we want to be able to like design something, test it quickly with picking up manipulating objects, and see how it performs and iterate quickly on that.

[00:03:57]And then once we've got a good idea of like how we want that to look, we kind of then do a more detailed design of the electronics and then make it a bit more bit more polished. So, this is this manipulator is obviously an impressive piece of the robot. It would be great to see the whole thing. Would it be possible to look at an entire KAI 1?

[00:04:16]Yeah, absolutely. Um I think Sam, my colleague, will introduce you to that.

[00:04:19]Great. Josh from the future here. I just wanted to quickly say that as you'd probably expect when chatting to a bunch of robotics engineers, things got pretty technical here. More so than I thought was suitable for this video. So, I'm going to be releasing a longer cut of this video allowing it to go in more into the technical aspects of the robotics, the software, and the business side of things. And that will be available to my channel members. So, if you get to the end of this video and decide you would like more of this Johnny 5 esque capade, hit the join button below any of my videos to become a channel member. And not only will you get access to this longer cut of the KAI video, you'll also get early access to my future videos, a discount on my store, and a bunch of other goodies, as well as supporting me making more videos going forward.

[00:05:04]Right.

[00:05:05]Back to KAI HQ.

[00:05:07]Hi Sam, it's great to actually see one of these in person. Can you tell me what it is you do here and how long you've been here at KAI?

[00:05:16]I'm a mechanical engineer at KAI.

[00:05:18]I joined about a year and a half ago. As a mechanical engineer, I'm responsible for designing the the body and the I guess the general hardware of the robot.

[00:05:28]Great. Well, that's perfect cuz I would love to get a bit of a tour of the hardware of this robot. What's actually running underneath this shelf? Uh we have the perception system here. We have a rotating chest.

[00:05:42]Then obviously you've got two arms, which are 7 degrees of freedom each. So, they've got several more joints than we've got ourselves.

[00:05:50]Exactly. And then got the torso, which is a 2° of freedom system, and that's mostly there to regulate the height of the what we call like the work platform.

[00:05:59]So, we design it to not be humanoid, but human-centric and to work in human-centric environments.

[00:06:06]So, you use the term human-centric rather than humanoid, which I think is a a great way to describe it because obviously, it doesn't have legs, so people may be a bit confused when you describe it as a humanoid robots. I'm presuming that this was for a reason you chose to go with a wheel-based rather than legs. Does this give some sort of like benefits or advantages? So, I can't take credit for the overall architecture. That was our CEO's kind of vision, and that's the reason why he convinced me to join the company.

[00:06:34]We both share the same idea that we both love humanoids, but neither of us believe that they're kind of going to be the the near immediate future. They're too complicated, too expensive, and in my opinion, most importantly, like more way more dangerous because they're not inherently stable.

[00:06:50]So, by having a a wheel base, which is actually from our CEO's former company, um it allows us to have like a inherently safe platform. And then everything else we build on top of that, we try to make it uh so it's, you know, as human-like as possible, but without any of the the cost or safety disadvantages of being a humanoid.

[00:07:10]Right. So, you're designing for function rather than flare. Exactly. Brian, like myself, wants to design a robot that can do what humans can, but not actually copying humans exactly.

[00:07:21]Uh which is why you know, we have um this shape here basically. In my experience, I don't think legs are necessary, at least for like the immediate future for factory work.

[00:07:32]Were there any particular parts of this robot that you found an interesting challenge when designing?

[00:07:39]A surprising amount of this is really printed, and we use lots of different techniques as well depending of the performance and cost. So, all the like goldish titanium links are made out of SLM aluminum.

[00:07:53]The the black bands and caps is a combination of SLS and MJF that we get made in the in the UK.

[00:08:00]The rubber bands you see here are actually um like vacuum cast from silicone rubber, and they cover like CNC parts so we don't have as many visible screws. It makes the design look a bit cleaner.

[00:08:12]Uh keeps a smaller joint as well. So, I'm guessing things like dust ingress and stuff isn't going to be an issue.

[00:08:17]Okay. And very often we uh use FDM to print the grippers. Because we've got lots of bamboos in the office, it makes it really easy to iterate uh very quickly. And the best thing about it is that if it breaks in the field or we need to make changes, we can quickly make parts here and then ship them out to the customer site uh basically next day. And because you can replace these hot swappable, they can just slap it in themselves. You don't need to send an engineer out there to do it for them.

[00:08:41]There is there a plan for future iterations of this where you'd be able to, for example, swap out an entire limb for a specific purpose? We can in fact do that actually. If underneath uh there's a rubber band, only two screws hold the arm on. The entire arm comes off in two screws. Amazing.

[00:09:01]So, you could create an entirely new appendage for a new uh for solving a new problem, ship that out to the customer, and they can effectively they just whack it on to their robot.

[00:09:12]>> Exactly. That's pretty cool. We tried to make it as modular as possible. So, the the head removes in one piece, the arms remove in one piece, the torso unbolts from the base, and then also our computer system is a separate module that can also be quickly removed and and upgraded. Well, this is obviously really impressive bit of kit, and you've obviously worked hard on making this look and function amazingly. But, without the software, this is effectively just a sexy paperweight. So, to find out more about how this works inside, we're now going to speak to Mark.

[00:09:47]Mark, uh you're head of robotics here at Kinze and basically in charge of making sure everything works together as a functional robot. I'd love to know more about how the KA1 experiences the world.

[00:10:01]I can see an array of sensors up here on the I'm going to first with the head unit and um what's it doing to understand the world around it? Yeah, great question. So, there there is a multitude of sensors.

[00:10:13]We have We have upward-facing cameras down here. We have lidar. We have another special camera here and then as you're uh correct in point about there's two uh two cameras in here.

[00:10:24]Uh it it does do RGB, but it also has depth. So, we also we perceive the world in things like point clouds and and depth images as well. Um and we also have um force torque sensors in the wrist. It really kind of depends on the task, but uh we obviously use RGB for doing lots of detection sort of stuff.

[00:10:42]So, saying uh is this a box? Where is a box? Um where where's the thing that we want to be able to pick up? But then, we can use things like the depth as well to determine like, "Okay, where what's what's a good grasp pose? Or does this align with what we already know as a heuristic maybe of like what this object should look like in the world? How is it How is it oriented?" And things like that. Um the force torque sensor thing in the in the wrist is really interesting and great cuz it gives us that kind of bit of feedback that humans actually have, you know, when you're trying to press or clamp down or push against something. So, it means that we don't break things. Uh you know, if a a good example would be if you're trying to pick up a a crate, like it's very easy for a robot to just kind of want to rip it apart and this allows us to be gentle and not break things. Um and it also means that going forward we can we can have kind of compliant behaviors. So, if I push the robot, then it will kind of go with it as well. Okay, so it's making them safer as well as being able to do their job better.

[00:11:42]So, when it comes to trying to teach the KR1 specific task, um how do you go about doing that? Do you need to sort of sit down and basically program in lift this object, put it there, or is there a more human-friendly way of doing it?

[00:11:57]Sure. So, there are a couple of ways in which we can do it. And again, it somewhat depends on the task. There are certain parts of the pipeline that we can do really well um uh using like more classical methods and like typical robotics algorithms.

[00:12:14]But, on the particularly difficult parts, we we have some world-class engineers and researchers here um with a lot of experience in in advanced AI methods such as imitation learning and reinforcement learning.

[00:12:27]These are really interesting because they don't require us doing a lot of uh kind of hand tuning or coding, etc. It's more getting the robot to actually learn itself how to do these things. So, for example, if we wanted to get the robot to learn how to pick and place this mug, it it would be much nicer to use these AI methods uh rather than you know, having to train something and get the perception stack to understand what the cup is, where's a good grasp point, manually defining sort of frames, and getting it to use all these algorithms to pick and place and determine where to put it. Instead, I can just show it exactly what I mean. I could get the robot to pick this up, put it down.

[00:13:07]Maybe we we want it robust, and maybe we want it robust in doing different cups in different locations. So, we'll we'll take a a fair bit of data and get it to do it multiple times. But then, once we push it through our our training algorithms, it will then learn how to do that itself, and we can just roll it out. So, it's a monkey see, monkey do type situation.

[00:13:28]It's really impressive seeing this thing sat here and sitting right next to it, but it's just a static object. Is it possible to see this in action? Of course, yeah. I mean, I think we're we're even collecting data today as well. So, if you fancy, come take it for a spin with us. Great. Sounds good.

[00:13:45]So, this is the uh KO1. What are we going to do with it?

[00:13:49]Um we're going to teleoperate it. Um so, Basically remote controlling.

[00:13:54]Pretty much, yeah. Um it's remote controlling the kind of movement of the arms.

[00:13:58]Um and it's kind of how we teach the robot to do things.

[00:14:01]>> Using imitation learning, which I spoke uh with Mark about earlier. So, I'm guessing I need to pop this on my head.

[00:14:06]>> pop that on.

[00:14:09]The way it's going to work is if you squeeze both triggers, that kind of connects you to the arms motion. Okay.

[00:14:15]And then if I let go, I can move and then readjust for when I Exactly. Okay.

[00:14:21]This is very cool. So, the task here, as I understand it, is I need to try and pick up one of these bottles. Yes, exactly. Have you got any tips?

[00:14:30]Um I'd come in from the top. Okay.

[00:14:34]Being in control of one of these curvaceous Johnny-5-esque robots was great fun.

[00:14:39]To control it, all you had to do was squeeze the two triggers on one controller, and whilst held down, that arm of the robot would mimic that hand's movement.

[00:14:49]Does this make me a roboticist?

[00:14:53]The speed that you can run the robots at is actually surprisingly fast, but it has a built-in collision avoidance, which prevents you being able to run one part of the robot into another part of the robot.

[00:15:04]And that's one bottle down. Whilst the KR1 isn't bipedal, it is bimanual. It has two arms and two hands. So, I thought I'd give the left arm a try.

[00:15:14]Being right-handed, this was a little bit more challenging, making sure I got the movements correct, but it was still surprisingly easy to do.

[00:15:22]Lifting and putting down a bottle isn't going to win me any awards, but it was fun.

[00:15:27]It's surprisingly intuitive to pick up on and to to operate it. Now it's see if it's time for these robots to start taking over other people's jobs.

[00:15:37]Don't worry, I wasn't suggesting that he replace you guys. I was suggesting that he had a go at being my cameraman and maybe replacing my camera operator Guthrie.

[00:15:46]So I thought as a first step, why not have the K R 1 have a stab at trying to do a humanoid robotics selfie?

[00:15:59]He's certainly not going to win any Oscars, so I think the camera operators out there are safe for now.

[00:16:05]Okay, and let's see if I can actually put this back down.

[00:16:15]Nope.

[00:16:20]Well, that was great fun. I could be doing that all day, but um I should probably let you guys get back to actually developing this thing. Uh thanks very much. It's been Thank you.

[00:16:29]really good fun. And I'm guessing as I'm leaving now, I get to take one of these home with me, yeah? Unfortunately not, but we have a 3D printed model. Oh, look at that guy.

[00:16:40]Oh, I like him. Uh do you think we'd be able to get a copy of that to share with the audience? Yeah, absolutely. Oh, amazing. Well, uh look in the description and I'll pop an STL for this guy down there. Thanks very much, guys. Thank you very much.

[00:16:52]Cheers.

[00:16:54]Whilst we aren't yet at the stage of having humanoid robots collecting our groceries or cleaning up after us at home, Brendan and his team here have proven that if you focus on function over flare, they are ready for the working world.

[00:17:06]This has been great fun to make this video and a massive thank you to Guthrie and the team here for making it all happen.

[00:17:13]I know this is very different to my norm. Um don't worry, there is a lot of 3D printing and um maker tech tutorials on the way. I'd love to hear what you thought of this and if you'd like to see more videos like this in the future. So, let me know in the comments. For now, I've got to go because I've had a notification that my robot vacuum at home has got stuck. So, I need to go rescue him. But, I'll leave you with this guy to sign you out.

[00:17:35]For now, thanks very much everyone and until next time, happy printing.

[00:17:41]I've pre-painted and sliced the printable version of the K 1 that Knesis gave me and I've linked to that in the description below.

[00:17:47]Just a reminder that if you would like a longer, more technical cut of my trip to Knesis, that video is available to all of my channel members. So, just hit join beneath any of my videos.

[00:17:56]I hope to bring out more videos like this in the future, but I've already got a ton of 3D printing and maker tech tutorials, deep dives, and reviews in progress. But, until they come out, why not chuck on one of my other videos to learn something new or just have some fun. Thanks very much everyone and until next time, happy printing.