Mia Hand: For Humans and Robots

[00:00:00]and in particular me hand. So we have built it around um a three actuation three actuators uh system. So the hand can perform the flexion extension of the thumb. Last three fingers are coupled together and then the flexion extension of the index finger that is also coupled with opposition of the thumb. So inside you have a special mechanism that is based on a Geneva drive and a four bar linkage that basically couples the position of the thumb with the when the index finger flexes so extends so that you don't you are not grasping something you basically switch position of the thumb so that you can switch between these two main positions. The simplification, I've said this many times, is that dexterity lies more in the head than it does in the hand. So if you have a system that's very smart, you can reduce the degrees of freedom and still accomplish quite a lot of dexterity.

[00:01:01]>> In prosthetics, we are more uh optimizing the hand. We have recently released the Mark2 version that has very easily replaceable fingers, a fabric glove that that is much more robust than u silicon glove. I mean gloves are rather huge issue for for hands in particular. And what about putting sensors actually in the gloves? I mean, that's one of the ways I want is does the touch sensing actually have to be part of the hand or is it is it easy to engineer maybe into the glove itself and that way could potentially be a consumable or disposable? I mean, it's like we don't want them to be too expensive, but is it easier to do that than routing it through the joints directly?

[00:02:10]So yeah, so Franchesco, welcome to the soft robotics podcast. Today we are going to talk a little bit about hands again and my understanding is that you've been developing hands both for um uh orthopedic kind of replacements for people who um who need to have an artificial hand and also for robotics.

[00:02:30]So there's obviously some crossover between the two and different requirements and needs but usually with cross-pollonization one side can learn from the other. So what did you learn about the challenges of trying to build a hand um for a human patient and also for a robot?

[00:02:49]>> Yeah, you are right. So at Princia we have this product that is called me hand that is here. I can also show you later and uh we have this in different versions. So one is for prosthetics and the other one is for uh robotics and industrial automation in general. So uh we at Princilla actually started developing robotic hands for research then moved to prosthetics and today we also expanded into robotics. What we have uh had to do in prosthetics is basically simplify a lot. Uh because uh in that um market you have the problem that people uh the human machine interface between the person and the device is limited in the no bandwidth of information that you can exchange. So that people can only control uh a few number of movements, degrees of freedom.

[00:03:45]uh in an intuitive way let's say so usually people in prosthetics do not control every single single finger movement but they basically select a specific grip and then the hand uh performs that grip uh in a predefined way so that for example I can show you um the hand inside has memorized the cylindrical grip that is the one that you use with large objects and then opening and closing your move inside of that grasp. Then you basically issue a special command. You switch to the pinch grip. That is the one that is used to grasp small objects and then you use that one or the lateral grip as an example. So you have the flexibility of switching between these grasps. There are different ways to do that. either through a special command that the person can issue or more advanced systems that use articular intelligence to understand the intent of the person and select that specific grip. But then while you are in a grip you basically on opening you have two um basic formats that you can send hand.

[00:05:02]Okay. And of course you also have a lot of restrictions on weight power consumption. So know there you have to try to optimize as much as much as possible in order to get the best performance in that case and in particular me hand. So we have built it around um a three actuation three actuators uh system. So the hand can perform the flexion extension of the thumb. Last three fingers are coupled together and then the flexion extension of the index finger that is also coupled with opposition of the thumb. So inside you have a special mechanism that is based on a Geneva drive and a four bar linkage that basically couples the position of the thumb with the when the index finger flexes so extends so that you don't you are not grasping something. You basically switch position of the thumb so that you can switch between these two main positions of thumb opposed versus thumb uh not opposed let's say that you can use to perform the uh the lateral grip and why is that I mean also in prosthetics for example there are let's say that the most advanced most dexterous prostthesis have six motors so one per finger and two for the thumb let's say So that's the gold standard but also in that case as I said people cannot really control over all of those degrees of freedom independently. Okay.

[00:06:41]So they have to fall back to a simpler solution, a simpler control let's say and uh so we said okay people cannot access that uh complexity so reduce that use less motors and this allows us to use larger motors uh where this can provide more torque more speed. So we have less grips but those grips have more performance with resp to the other ones. And uh so how we did select those specific grasps?

[00:07:17]We went into the literature and looking at uh grasp taxonomies that is basically a way to classify uh the grasps that humans do during activities of daily living and measuring frequencies of occurrence and basically saying okay these are the grasps that are used the most. Let's cover the uh movements that allow you to perform most of them. And we basically selected the movements that uh cover seven out of the 10 most used grasps during activities of daily living by people.

[00:07:52]And this basically covers like 85% of the occurrences as a frequency of the specific grasps that are used during uh daily life. And these works also were done through different uh uh professions like households versus workshops. So that know you can balance a little bit what people at home do versus people uh professionals do that work with their own hands. Basically >> this is the general framework that we started with prosthetics that we then basically also moved and translated into robotics today.

[00:08:38]Um so yeah that it's really interesting the challenges that you have there. Of course um it's a medical device so you you have to follow a lot of guidelines there. I mean the most important thing is it has to be comfortable. Uh and then you also have to make sure it's something that uh can be easily cleaned and everything else. So usually components will be made out of like titanium or they'll be made out of carbon composite to make them light and everything else. And then there of course the interface you the ideal thing is that you have some sort of you know BCI or brain control interface where a person can just think about what they want to do and you have full control.

[00:09:12]But we're not there yet. In many cases, it's it's a it's a very simple kind of movements of muscles that you're reading that and from that trying to gather a little bit of the intent and then again sometimes the intent can be very explicit because you've got a couple of options on there that the uh the individual can just push exactly what they want to do and then it will kind of go into that particular mode. I'm assuming now with a lot of speech recognition, you could potentially get around actually needing to have a button interface and if a person can't think it what they want to do, maybe they could sort of say, "Oh, I want to do a pinch right now." And then it could go into that mode and then you would go ahead and activate it. Is is that a possible way of doing an interface?

[00:09:57]>> Yeah. Yeah. I mean, especially in research, people have been trying so many things. I would say that that that as as you said correctly the standard clinical uh system that is used today and that was used since like 40 years is EMG control. So recording the electrical activity from the residual muscles in the forearm and use that to control the prosttheesis.

[00:10:21]>> Yeah.

[00:10:21]>> The standard interfaces with two electrodes. So one for opening and one for closing that is the simplest ones.

[00:10:29]Then we have uh companies that are specialized in control system for prosthetics that develop machine learning uh based devices that allow to uh record up to eight signals and then decode them directly into an output that is the the wanted grasp.

[00:10:51]That's but also in that case you have like >> three to five movements that you can uh um decode.

[00:10:59]>> Okay.

[00:11:00]>> Yeah. So you could use speech recognition. It's not super uh use because people have to speak to the prosthesis. So it's not know when you imagine you are in a crowded place or somewhere you don't want to be talking to your hand to to make it do stuff. But uh yeah the the other option is invasive interfaces. No that would be in the brain but also in the periphery. So when you had electrodes implanted in your forearm and that allows for better uh cross lower cross talk to talk to uh you can have more signals uh and that allows you to control more degrees of freedom basically >> right and and you get kind of more of the control that way because this is a problem with a lot of robotic hands is that um there's similar design challenges there is that some cases it seems like robotic hands may be getting overengineered. They're trying to do way more than you really can do to try to cram that all in there. Whereas you've been forced with prosthetics that we got to get something that's usable, but we unfortunately we cannot restore all the degrees of freedom. We can get a certain amount that's in there. And of course, how much of that control do we have up in the brain? you know, humans were pretty much able to control everything, but we have a certain amount of reflex action there as well, but you're moving more of it down there, which we're probably going to see with a lot of robotic hands is again a lot of that compute you can kind of put down there.

[00:12:26]So, it's more of a high level command of grasp and then a grasping. Now, not all grasp are the same. A lot of times everyone just thinks mechanism close.

[00:12:33]That's the old school way. It just kind of goes down and maybe hits some hard limit or something like that. But you can have much smarter grasp that depending upon the object that you're grasping, it knows that that's the ideal sequence. That's the way we need to do it, how hard we need to grasp, exactly what the configuration of the fingers, that's something that could be done very locally, very easily. And that means all you have to do is give that high level command. Um, again, the simplification, I've said this many times, is that dexterity lies more in the head than it does in the hand. So if you have a system that's very smart, you can reduce the degrees of freedom and still accomplish quite a lot of dexterity and and that's what you're finding. And hopefully it comes over to robotic applications that rather than trying to be fully general purpose, which is what everyone wants to do. I want the hand that's like super dextrous and can do absolutely everything. And as soon as you do that, you can't do anything because it's just too hard to get it to work. And if you kind of narrow your scope a little bit and say we need to have something that's functional and usable, you can get in there. So, it seems like you're you're learning that because if we want a robotic hand to work in like in the household, you have to categorize like what are the typical um gestures you need to do to get through the day and just limit it to that. So, it seems like you already know what that list is and maybe some roboticist should look at the list as well.

[00:13:54]>> Oh, I agree 100%. So I I think that in the present times the fact that uh AI exploded of course provided a new tool to control much more complex systems robotic systems and this tool was not available before but uh I have to say that uh on the other side we didn't have a hardware revolution no so robotic hands that are let's say copying the human hand have been around 20 to 30 years up to now. But the problem in making these robotic hands deployable is the fact that they eat a lot. They cost a lot. They are very very complex. So they break relatively often.

[00:14:38]So it is very difficult to I mean translate that kind of engineering is in something that is then deployable and useful. Mhm.

[00:14:48]>> So, so it's relatively simple to just copy the architecture of the human hand and say okay I have an arbor that is uh capable of doing similar same amount of movements but that that comes to a cost of uh associated with other components of the spec sheet of that again it's either weight or complexity or cost. So this is a little bit the um the balance that we are trying to find and I think this is the the real work of engineering no because uh not just copying but but finding the solution that is the best for the job that you have to do. So I personally do not really believe that we have to necessarily go towards a solution that is 100% general. So we can have specialized devices and grippers and hand of arm tools for specific tasks and I think that we can go on with uh simpler solutions for 90% of the activities that are out there and then if we really need to do very specific stuff then we can figure out a special tool for that.

[00:16:09]>> Yeah. Yeah. And it's amazing how adaptable um people can be when they they've lost that function and then you kind of restore it a little bit that again it comes down to sort of the the intelligence behind the the equipment that that you have that allows you to do a lot more than sometimes the designers originally thought were possible and figuring those those two things out to get the optimum. So you've gone down to three actuators, uh, which is probably like the the bare minimum that you you would want to use even for like a robotic arm. And are you finding that when you take your designs and now say, "All right, let's move this over to a robotic application that the engineers are able to get quite a lot of functionality out of it or do they still think, ah, I need a couple more degrees of freedom and and how do you kind of do the push back?" You know, think about is that now a lot of the engineers will say I give me more give me more and maybe you say well wait a minute why don't you just try working with what you have at first and and see if you can solve it that way.

[00:17:14]>> Yeah. So uh when you move to robotics you have different constraints you know as you also said before. So of course we the device is not 100% the same. So we added more sensors for sensors specifically in the thumb index and in the finger that you can use um with respect to the prosthesis. Uh in addition to that of course the number of degrees of freedom is a central discussion point. I would say one of the main limitations of the hand that I have to recognize is the is the fact that the thumb only has two positions. No. So you only can have like closed or open. And this is a little bit limiting because uh I mean you lose a little bit of flexibility even though um probably you don't use this degree of freedom in a in a very um fine way while grasping objects. So maybe if you would have like four to five positions you would be okay. That that's my point of view. And the on the other side you what we see as a limitation of this particular device is the fact that you only have one joint per finger. So for instance this can be limiting when you >> close the fingers in this way. So you do a cylindrical grip and of course if you have a very small object that you want to grasp you're not able to hold it steadily. So we lose a little bit of form factor there. So old in a object because of the form closure of the grasp rather than only pure friction grasp and this is something that both actually limitations are something that we are solving at the moment.

[00:19:06]>> Mhm.

[00:19:08]>> based on that uh there are a bunch of uh studies that basically study the human hand and basically tell you that okay you have three main functional components. One is the thumb, one is the index finger, and then the other one is the other fingers, let's say. And this Yeah. So, you don't really use a lot of mobility of the last three fingers coupled together. And this is also >> is there are some um data capture gloves out there, some UMI gloves that are are basically designed around that idea is that you get a thumb and index finger and then the other three that are basically taped together.

[00:19:44]>> Yeah.

[00:19:44]>> Go ahead.

[00:19:45]>> Yeah. Yeah. And also you see um a lot of grippers that are three-finger grippers, you know. So >> why is that? Because basically that's the minimum amount that you need in order to have some flexibility and move outside of the pure pinch grip. That is the one that uh uh grippers are uh >> and well also comes down to just this pure stability. is the tripod grip and and the fact that you know everything is basically built out of triangles when you have stability and it's why you know stools have three legs on them and not two as soon as you get three you're not without it falling over and the fourth is like additional redundancy but you don't necessarily need the fourth leg so it's the same thing with a lot of pinch grips but the the other one might give you a bit more stability and strength in some places but you're absolutely right is that going with three allows you to do a majority of everything that you need. The other ones are nice to haves but not necessary.

[00:20:48]>> Yeah. Yeah. I think that also everyone is is as focused on that as a you know um as a first approach because of course you want to have endless possibilities with um a lot of degrees of freedom. But then for instance we were at IIKRA two weeks ago in Vienna and we had a conversation with a lot of researchers as well that they said okay um we we went in the past year with very high degrees of freedom hands but then these are not super reliable. We need then something that we can deploy and use more consistently and this is why we are looking at alternatives because of course you can overengineers everything but then at some point you need something that you know can be deployed.

[00:21:36]>> Yeah it's a funny pendulum swing. First it was like everyone's like let's reduce the number of degrees of freedom. Oh it's not enough. And then it's like came way over here. It's like let's give all the degrees of freedom the human hand has. And then they're beginning to realize man that's hard. How do you get control and strength of your grip and everything else? And we're running running out of room here and all that.

[00:21:53]So, it's good to see the pendulum is kind of swinging a bit more in the middle. And and that's sort of my feeling is that if you want to have the robots be able to already do many applications in industrial settings, you can do it with a much fewer set of degrees of freedom. And it comes down to some a very simple premise. And that is majority of the jobs we have out there.

[00:22:15]The assumption is that the people we're hiring don't know how to play the piano or maybe even the guitar. You know, it's it's not part of the so so we don't need that level of dexterity. And the other thing is that a lot of times when people are working, they're having to wear gloves and everything else which also reduces kind of the degrees of freedom that they have there. And so it's like, well, wait a minute. All these industrial applications are literally kind of like dumbing the hand down to like the most basic functions. Maybe we should think about designing our our robotic hands that way. And so not only would they be cheaper, but they'll probably be more robust because, you know, every every time you add one more part, that's one more point of failure.

[00:22:53]And the the more that you can kind of reduce, not only can you make those remaining parts more robust, but then there's less chance of something going wrong.

[00:23:02]>> Yeah. Yeah. You're right. I mean gloves for instance limit mobility but also sensitivity you know so also a lot of jobs do do not really need a thousand sensitive points on the fingers in order to be done and I think that the main point here is the fact that in the industry it's true that we want to help humans and adapt the robots to do the things that humans do. So we want their flexibility etc. But on the other side we had to think keep in mind that if we want to deploy something then the real benchmark is the gripper because the gripper is the one that today is deployed and that it works. So when you're going to talk to manufacturers then they say okay I have a gripper what can you give me more than that as an alternative? No. So this is uh this is the conversation that you have to keep going. And of course uh one big point is reliability. So you can put a lot of stuff in your fingers let's say. So if you want to have a 20 degrees of freedom hand, you can one of the approaches is to have very small motors in the fingers as an example in the joint. That of course on one side simplifies the mechanical transmission because you have quasi basically direct drive between the motor and the joint. But on the other side if one finger breaks then you are losing like four motors inside of the finger and and that's going to be expensive. No. So you want to limit the complexity of the fingers because those are the weak point of your device. Uh trying to keep know flexibility as as high as possible. So this is also one of the reasons why we try to keep everything inside of the of the palm of the hand in order to uh be be able to manage finger fingers breaking because they can you can just replace them very quickly rather than uh the need to know rewire everything inside of the finger because you have four motors then you have to repass wires through the the hand and that's going to uh be much more difficult to do also from a customer point of Yeah. Yeah. The the more wires you have to route, the more likely that they're going to fail on you. And I I just want to touch a little bit on the idea of that, you know, the different modes of sensing. So, uh the robotic hand, you you have some touch sensing in there to to help the the overall functionality of it. But it's I guess with a prosthetic is a little bit different. you probably have some touch sensing that um the fact that you're kind of running a program on on there on a grip that it needs to know when it's coming in contact with something. But at the same time, it's really tricky to give that feedback to the patient because there's no really good way to get the kind of haptics. But at the same time, it is mounted and you get a certain amount of propriception that kind of comes from it. So if the hand is in contact, they are they're feeling it through the interface and at some point even though there isn't really a sensor there, if you're good enough with it, you just can kind of feel that it's there, right?

[00:26:15]It's it's amazing how the how patients will get that kind of idea. And it's akin to when we drive our cars, you you swear that you can almost feel the road surface through the steering wheel as as if you actually had something on your tires. And of course you know you don't the same thing when we grab tools and everything like that is we can feel a lot of things like that. So imagine even though you don't have a sensor on it there's something about human anatomy that we very quickly are able to dial in and know when that point of contact is.

[00:26:45]So while it would be nice to have that it looks like from the standpoint if you look kind of at the first principles people are able to adapt well enough that I almost wonder it's like why can't we do it with the robots? Do do we really need to have sensing or could we do it through just kind of this force feedback probe? Now the thing is we are sensing it in so many different places when you mount it up here that we maybe have enough information versus if you just have you know a simple mounted interface at at the end of the robot arm maybe there's not quite enough information there so we have to go to the next step. But what are your thoughts? How much do you think you could do it just through prop reception feedback versus actually having to have touch sensing?

[00:27:26]Yeah. So what you describe is called um um incidental feedback. So basically feedback that people have but it's not designed to to have like noise from the motors like uh um the amount of uh force that they need to uh contract their muscles in order to activate the prostthesis and the vibration that is transmitted by from the hand to the socket. So it's all of those information that somehow is transferred also. It's not really the system designed for that.

[00:28:03]No, and this is something that we could exploit also in robots uh by placing fence or somewhere else not exactly in the hand but somewhere else let's say in the forearm and again this would also uh help in u uh kind of managing the complexity no not putting all of the complexity inside of the end itself but distributing it a little bit. So this is something that could be done. Another thing that can be done is for instance there are a lot of studies where uh that study human endover. So passing human robot and so passing objects between humans and robots and researchers have basically seen that a lot of information does not come from the uh fingertips but comes from the wrist. So you have to measure the interaction forces between the human and the robot in in order to be able to fluently exchange objects, but you can rely on a priest mounted load cell rather than having fingertip force sensors for that.

[00:29:14]>> Exactly. And in DLR, >> yeah, DLR showed something like that. I think it was back in February. Um they they posted something where they're taking their cobot arms and all they have is is a force torque sensors in the joints yet they can tell where you're touching on the robot arm and they can literally like spell out someone's name on the upper arm and there's no touch sensing there but just the forces from it and it took a lot of training and they do it with reasonably high accuracy and that already to me was telling me it's like yeah it's just if you get enough of that training data and you kind of know what's going on with your model eventually you can figure out where a force is being applied uh and and how now I think in the case of DLR it was like it was sort of like one point force if you had like multiple points of contact it gets really tricky because then you're getting too much signal and try how you'd be able to extract it but again it's kind of this real proof of principle that um there's way more information out there I'm aware of and if you know your kinematic structure in that for some reason our own neural networks internally are very good at learning that really quickly And maybe part of it is because we also couple it a little bit with vision. So we we we are sort of associating the fact that oh I know the finger is touching there. I know what it feels like when it's there. So if you close your eyes and you you feel some kind of load again, I know exactly what's going on. Yeah.

[00:30:36]>> Yeah. It's also another point that we want to point out is the fact that humans do not really rely continuously on closed loop feedback. So what when I need to grasp an object and lift it, I'm not saying okay now I am applying two newtons of force that's enough for this object. I'm going to lift it. Let's see if it's slipping or not. But basically we do a lot of our tasks in a feed forward way because it's it's >> much faster. Okay. So what you do, what happens when you lift a a bottle of water that you think it's full, but then it's half empty, you overshoot. And the fact that you overshoot then triggers a feedback system that basically allows you to adapt. But only in that case, you're using that no tactile information. And also this feed forward versus feedback system is something that you can exploit to basically in order to uh get the bandwidth of the information between the hand and the system down uh and be faster. Okay.

[00:31:44]>> Yeah. Yeah. And you that's sort of an interesting observation of how we are a little bit different in that um yeah engineering we we are applying mathematical formulas uh to solve a lot of our problem when the reality is we don't do that. [laughter] we have this innate sense how things work and that we you know we we have a world model which of course is what everyone is trying to do right now um with with the robots where we already we know what the consequences are of our actions and how everything should should behave whereas when you try to do the mathematical modeling it's um it's a little bit trickier to get the same results. So, so in many cases we are trying to solve the same problem with radically different approaches which is why we should not be surprised that we get different results.

[00:32:34]>> Yeah. But this also means that for instance you could have a learning phase where you have a hand a device that is more complex with a bunch of sensors in order to learn the world model basically and then when you deploy the system you have a simpler system that does not have that amount of sensitivity and complexity but then can still trigger correct corrective actions if the model diverges too much from from the current obser observation. So the the sensing that you get. Okay. So this is also uh for us a way to think about force sensing because uh uh sensors are something that uh can also break. So you don't want to overengineer that as well.

[00:33:24]And also putting them at the fingertip is and can be tricky because the fingertip is really the interface between the object and the hand. So is this the part that is going to wear out uh first? Oh, because of friction of course.

[00:33:39]>> So yeah.

[00:33:42]>> Yeah, that that is Yeah. And that is one of the biggest problems with with touch sensing is that not only is it just difficult to get the sensors uh to have the sensitivity that you want, but they are literally where the kind of like you know the rubber meets the road. So that that's where the robustness has to be, but it also has to be sensitive. And then when you collect all that data, you got to figure out, well, how do I transmit it from my fingertips all the way to where my processing is. I have all these joints I got to be moving everything through, which kind of complicates it, which means that's one more thing that could potentially fail.

[00:34:15]So again, it's that balance between um wanting to have a lot of data, but at the same time, the complexity it's going to bring into your overall system that might make it far less robust and far less capable. and and the whole idea of optimization is figuring out what's the bare minimum I need to really be able to accomplish a rich variety of different tasks.

[00:34:39]>> Yes. Yes. Correct. And also say and another point that I can say about know this complexity thing is uh the fact that another important point for uh industrial deployment is about IP so ingress protection. So this is something that we are uh actively researching for and this becomes more and more difficult more and more stuff you put in inside of your fingers you know because the finger is the extremity and you have a lot of joints hovering those joints in order to allow them to be basically waterproof is very difficult and uh so having an architecture that allows you to keep the fingers let's say empty from a electronics point of view also allows simplifies you the way towards higher IP uh levels.

[00:35:37]>> Yeah. Like like I say the the IP at your IP uh requires a lot of IP. So ba basically the intrusion protection at those interfolial joints that that you have there uh is quite I guess you say an intellectual property issue. So hence the the three IPs you're gonna solve at the same time. Yeah, I mean is it's super challenging because it's robustness means a lot of different things. Uh as well as safety and um just you know just the fact you get movement those joints open up. It's very hard to design something where you're not going to have something open up and then you need to kind of close it off and then the closing of that complicates everything. And then again depending upon what your actuation strategy is that may be a lot harder for you to mitigate overheating. So if you have the actuators actually at the joints, well there's a lot of heat you got to get rid of. How you going to do it if you're going to be putting a thick glove over that you need to conduct that away. So yeah, it's it's a huge challenge. Um so going forward, what are the improvements you hope to be uh to making to these for for both as a prosthetic and also as um a robot hand and where can everyone find more information about both?

[00:36:53]Yeah. So we we are working actively at the moment on both sides. I would say that in uh in robotics in prosthetics we are more uh optimizing the hand. We are recently released the the mark 2 version that has very easily replaceable fingers. A fabric glove that that is much more robust than u silicon glove. I mean gloves are another huge issue for for hands in particular.

[00:37:23]>> And what about for putting sensors actually in the gloves? I mean that's what I always want is does the touch sensing actually have to be part of the hand or is it is it easy to engineer maybe into the glove itself and that way could potentially be a consumable or disposable. I mean it's like we don't want them to be too expensive but is it easier to do that than routing it through the joints directly?

[00:37:47]Uh I don't know. Uh I think that you can do that but you would it depends on the sensitivity and on the let's say um quality of the force information that you need from that sensor because if you need it to be uh disposable then basically means that to cost uh very little. So probably the quality of the you you probably are not going to get quantitative information about the force but maybe more qualitative information. So uh like low, medium and high forces rather than you know an exact amount. So it might be something that you can do uh depending on the scenario maybe that's that's enough. So yeah that's one way to go. The problem with the globe is that itself is uh uh something that you have to replace very often. So I'm not sure if you want to put additional uh let's say uh components there because this means that you are increasing the cost of the component that you have to replace.

[00:38:54]>> Yeah.

[00:38:56]Okay. So yeah, I mean that that's that's always the the trade-off between um the cost of what's going to wear out and you know serviceability and and everything else. Um yeah, so what what other are the kind of the big challenges that you hope to be able to resolve in the coming months?

[00:39:16]>> So we we are focusing a lot on uh on the as I said also in the industrial version of our products. we are developing uh new stuff that basically uh go um try to resolve the issues that mentioned before. So a little bit more dexterity but uh staying always in the that kind of range in the number of degrees number of actuations you of actuators >> but increasing the number of degrees of freedom. So number of joints adding a little bit of also mechanical intelligence and under actuation and this allows to have also kind of adaptability of the grasp to the object without losing precision. So we are trying to avoid also to use uh tendon driven actuation systems because of the reliability. So trying to stay in the gears and links uh transmission systems and uh going up with the ingress protection because we think that uh >> that's one of the I mean main points. So you you don't there is no hand [laughter] there is no hand on the market that has uh IP protection. you know uh rather than having just a hand with a glove on top of it which is of course you know easy solution but not reliable I would say and we are trying to push that uh in order to allow also those kind of applications that need that uh that requirement >> yeah and and that's that's what I've seen everyone is trying to solve that IP problem with the glove and when you go over the joint you get that big opening and the glove is stretching over it and then when finger comes up, I'm it's like, "Oh, I bet that gets in there and pinches because it's, you know, it's an ingress point, you know, and it so after amount of time, you could just imagine that it's going to chew right through the material and that's why you have to replace the gloves. So, a lot of it is coming up with the design that it's already um, you know, eliminates the problem in the first place, but there's always probably going to have to be some covering. And again, so for some of the touches, you know, everyone keeps on talking about being able to uh do, you know, uh slip detection. So you you need to have the the sensors that are able to tell when something is sliding out because that's very important to your overall grasping. It's not that I want to squeeze the bottle until the pot comes up. I just want to I just want to hold it enough that that doesn't happen.

[00:41:47]And and that that's something that you want to be a local feedback loop that I I shouldn't even have to think about or be conscious of. It's just like grab this thing and make sure that the grip is proper. It's not too strong and it's not too weak and a lot of times it means you use slip protection. Uh and so yeah, I guess that's like are you you working on getting the sensors to be able to to work at that level?

[00:42:09]>> Yeah. Yeah. So we we have recently partnered with Touch Lab UK company that have developed I mean a lot of for sensors. They also developed custom sensor for for the MIA hand. It's not this version but it's covering the tambber the sensor with six 16 tactile points and uh so we have partnering with them because uh uh with so the with that sensor we are able to we can detect sleep uh with that kind of technology. So yeah so this is something that we are also looking for. Good, good, good. Um, yeah.

[00:42:48]Mara, do you have any questions?

[00:42:50]>> No, I think you covered everything. It was great. Yeah.

[00:42:53]>> Okay. So, I guess, uh, you know, an interesting thing is that, um, you were located, um, in the part of Italy that I think most people are kind of familiar with or have heard of, and that is you're close to Giza, correct? And so, yes, >> we all know about some famous tower that that's there. Um, it's um, it's a lovely area. I've uh I've visited there but it was about a decade ago and uh yeah so any anyone >> you're well welcome to come back and we will be pleasured to to have you here.

[00:43:22]>> Absolutely. Like you say it's it's it's been a while but you know I always enjoy the trips to that part of Italy. It's like it's the food is great the people is great interesting technology. A lot of the reasons why I was around there is because working with um some pretty smart Italian firms and individuals was always an excuse to come down there. So once again, uh it's a pleasure having you on there. Any any last thoughts?

[00:43:46]>> Oh, really thank you very much for for the opportunity and for the discussion.

[00:43:49]It's it's been a pleasure for me and uh really looking forward to meet you in person and uh have a additional discussion.

[00:43:58]>> Yeah. Yeah. And and and have a gelato.

[00:44:01][laughter] >> Why not?

[00:44:03]>> Absolutely. Okay. Thank you, Franchesco.

[00:44:06]And so so ciao.