Interview with Arash Ajoudani

Q1: So it would be wonderful if you could start with giving us your name and tell us a little bit about what you worked for in your early years.
Arash Ajoudani: Absolutely. So my name is Arash Ajoudani. I am the director of research at Atari two labs in Italy. This sort of technology. We are in Genova. I was originally born in Tabriz, in Iran, but I did my PhD in Italy. So in 2015, I moved to LA and I started doing my PhD at University of Pisa. But it was a joint program between the Italian Institute of Technology and University of Pisa when I enrolled.
I did a little bit of internship before to understand if this is the right topic again and if I get along with my supervisors, which, where Antonio became Nico Segarra. They are great colleagues now. We are doing a lot of projects together. And, this actually attracted me a lot. So I was looking into expanding research, about my knowledge of art in general.
Human robot interaction, because I come from biomedical engineering and controlled backgrounds. So I did my master’s in control theory, where my bachelor’s was in biomedical engineering, so I wanted to. Yeah, kind of, you know, implement and find a synergy between the two. And when I thought about it and I. Human robot interaction came really, really, really fit, I said, I can bring some knowledge about humans.
I can bring some knowledge about control, digitization of, of robots, of actions of robots. And I can connect them together. In fact, my PhD started to look into some new forms of teleoperation of robots. We looked into how we can tell you operate robots, not only their position, their velocity, but we looked into their how their if physical interaction characteristics can change, like the way impedance, you know, humans for instance, when deal with environments because our muscles to be stiffer and softer when we are dealing with uncertainty with with different dynamics of the, of the, of the environment.
So we tried to replicate that through a teleoperation interface. And we managed this. We called it tele impedance. And it started a really nice line of research actually a few maybe a year ago, we published a paper like ten years of telling PE dance like from the day we started that, and then we started bringing it towards the end of PhD, like looking into, you know, bringing in general human interaction, physical interaction capacity to different types of robots, like collaborative robots was that was the where the main focus we started looking to exoskeletons, we started looking in the hands.
So a lot of hardware, but the principles were the same. Let’s try to replicate human physical interaction skills, which makes us really efficient in dealing with environments that we have never been to. We did that and I finished my PhD. Actually, my Ph.D. was awarded among the top three, best European thesis, George Roth Award finalist, of Europe.
And, I did a little bit of postdoc, always at the Italian sort of technology. And after a few months, I kind of applied to this tenure track position at I-80 because I was advised and suggested to do so, and I applied for that and there were two categories: senior category and junior category. And I wrote in the junior category, there was a big competition.
And, and and I’m so lucky and, that I have wanted since 2015, I started my lab. We call it human robot entropy, human robot interfaces and interaction, and, just a year after a few, maybe a year and a half after I won a New York grant and I started coding a large proposals like Sophie I’s one of the most known in the European Community because it was awarded, it was one of the most successful European projects back then, in 2024, when we when we concluded it.
And, this boosted not only the team capacity, but also the knowledge we wanted to create in our power, in our lab. So we started the lab looking into how we can make robots and humans very intuitively collaborate, not pushed. And by not pushing, it created a very complex, let’s say, chain of requirements.
On one hand, you need to know exactly what a human is doing. You know, interpret the actions, interpret the states that the robot should know, but not everybody can understand by looking at you. And this is states where simple movements from simple movements to be reactive to more complicated ones, like a physical state of a person. In terms of ergonomics, if you’re working with a robot.
So what’s the best way of collaborating with a person, handing over an object or or doing some work together that makes sure you’re physically organized? Programs are optimal, so the best posture or working condition makes sure that you are perfectly fit to that particular task. This requirement came from one of the, let’s say, issues that you’re facing in Europe and all the big industrial countries at the moment.
For instance, today, Europe is spending €240 billion a year wasted money associated with ergonomics, muscle, skeletal disorders and all this physical concerns that actually people get hurt and the insurance and the time waste and the productivity change. This costs a lot. Hero. And then we thought, okay, we need some projects to look into this more deeply.
There were some industrial, let’s say, sort of, evaluation methods, but they are all pen and paper based, question questionnaires based that they are not very efficient. They’re not customizable or personalized. And some of them are forced if you go to a worker and say, do you feel this work is heavy for you? Most of them are afraid of telling the truth.
So we thought of let’s build intelligent sensory systems now with AI algorithms or some model based. So hybrid approaches. By looking at the movements we can tell how the body actually states. The kinematics and dynamics of the body are changing. While we were doing this, we did a bunch of projects. We won a lot of awards. My first DRC five years was on that.
Several projects were done that we had several best paper awards, and they also brought humans and robots together. So now that you have a nice physical model of a person, the robot knows how to behave. The robot knows how to adapt its behavior, so the human is always in its best posture. Depending on the internal loads, external load imposed on the body, vibration, transmission from the toes.
So there’s always one optimal posture or condition of the work that minimizes the risk for you. People are struggling with this. We were happy with the progress, but at some point we thought about fixing the problem of human robot interaction. Then in the lab we said we never tackled the psychological aspects at the moment.
Actually, the psychological issues related to work are three times worse. So it’s about 600 billion euros. Yeah, in Europe and worldwide it will be more significant. And this is because of the anxiety, because the technology is imposing, the stress of work is imposing. And these are making people become more and more depressed, more and more anxious.
And this has a long lasting effect on people’s health. And this is less seen and less mediated compared to the physical one, because everybody thinks heavy load, you can reduce it. But what do we do with the anxiety, with the stress, with the workload, with the cognitive and mental demands of work? And there was, again, some really lab equipment.
So you put sensors in the brains of people, but this is not going to go to industry because you can’t put sensors on people. And there are some questionnaires. Again, there’s no way you can actually just move on your own. And then we started doing some projects on that. We started looking into body language. This was the first time we started looking at taking a step back, looking at the psychological aspects of movement, saying what happens if somebody is distressed?
When somebody is distressed, you start moving your eyes touching your face. You know, your attention changes. And all of these are really nice indicative measures of how your stress cognitive load is changing. And now we have really nice AI tools for perception using cameras. We could capture them, quantify them, and look at their variations and tell how the person in stressful robots could interpret people’s distress.
And when you’re stressed, they could just slow down, change the behavior, and mitigate your system. So we have papers that these days show that people are working with a robot, and the robot is optimizing your physical performance and your cognitive performance. But if you have signs of stress, the robot slows down until it engages you just right enough that you’re not bored, but you’re not overstressed as well.
So this adaptation across physical and cognitive factors, where the topic of my research for over ten years, which is continuing now, I will tell you a little bit about that, but it was the first time that robots actually were not only adapting to kinematics, which means that I’m here, move here. Robots were actually helping me to do this, considering beyond what the task and performance of the task is, that what it’s like mostly on the states of the robot with these contributions a few years back, I received the, Early Career Award of the Robotics and Automation Society for contributions to ergonomics, human robot interaction, and obviously also some projects awards here and there.
I got two ERC projects and a new, Italian ERC, which is again five years with some funding involved, which is called epic, the project. And in this project, we want to now look into the interconnection between physical and cognitive factors, always from a robotics perspective. Right. Because our physical and the mind and bodies are connected, when you’re tired, you make more mistakes physically and vice versa.
You know, you’re when you are physically tired, at the same time, you get cognitively more loaded, and easily stressed. So we’re trying to see if we can look at it as a, as a, as an interconnected complex problem. So how to make sure that we are dealing with both at the right moment. So the next five years from this May actually 2026 people look into this problem.
But robotics is the core element for us because robotics is not only a measure of mitigate it by looking into what is happening, but also mitigating that. Right? So if you have a worker, a human being, a humanoid, a collaborative robot, an exoskeleton that knows your states and always adapts to you looking into interconnection, or they are separate influences on your body and mind, this will be the best we can call him.
Everybody interacts. So we brought this concept beyond what physical or social human robot interaction was. We made the robots try to make them robots with a lot of contributions from society, from great colleagues. We have it and IEEE and other societies to make sure that robots are well aware of it, well aware of this fact along this path.
Obviously, the verticals, a lot of technologies and ideas, for instance, one of the requirements obviously, is human tracking. You need to be able to see what the human is doing, like the posture, the body language models and stuff like that. Back then when we started, only wearable systems were available. So you would put markers and cameras around, and then they would look into it through infrared.
And I remember wearing those suits and putting these markers would take hours. And every time you were moving fast, some of those markers would fail. And then the IMU based systems still came as a suit. You wear it, you wear them and stuff like that. But then Covid arrived. You were not really happy to move from one person to the other one.
And there was a hot day. People were you know, wearing the system and we were not comfortable passing from one person to the other one. And each suit costed a lot. And you said, you know what, probably we need to move towards market-less systems, market-less systems are the ones that you don’t wear anything, and systems will give you the same amount of information that you need.
We were dealing with this and actually some confirmation from industry came. I was visiting industrial partners in different projects and surprisingly, I was in one of the steel manufacturing kind of industries. And we entered the workspace and I noticed people dealing with sharp objects. They’re not wearing gloves, and some of them are not even wearing protective glasses.
And I asked them, how is it even possible? That should be said, workers do not really like to wear anything. You can build the most beautiful technology, but imagine yourself wearing eight hours of this kind of stuff, right? So you reduce mobility and so on. And this was another confirmation that marker-less tracking systems. Then we started.
This was the ERC proof of concept project awarded by the European Commission. We worked on a modular marker system, which is multiple cameras with an AI system all custom based with us from our team that you set up. Number of cameras the way you want, you cover the area and then it gives you multiple and multi-object tracking in real time.
This is actually becoming a startup idea, and hopefully we are kicking it out, and kicking it off. And, and a lot of people have been working on it and contributing to that will bring this into the society. And we use the system not only for robotics and automation and economics, and recently we are using it also for healthcare.
So thanks to this system, you can actually just look at people without really forcing them to wear anything. So we looked into the impairments of people while walking. So when people walk and sometimes do you see some asymmetry in their walking, some level of impairment can be, can be, you know, can be seen like trunk flexion. Some people have knee x, you know, extreme flexion and a lot of pathologies that are known to the, to the, to the biomechanics community.
We manage AI algorithms. Try to detect them, measure the distance from that to a normative personalized model. So if you were with your skeletal body and with your conditions, you should have walked like this and how far it is from the nominal condition. And then we tried automatically to identify where the issue is coming from.
So this tracking system goes beyond human robot interaction, goes to health care. We looked into sports. So automatic understanding of the number of you know mistakes. You do to avoid injuries or a trainer. Maybe just a camera or even one camera media application. It tells you what to do, what’s the mistake and how you do it better health care is huge.
So the impact is huge in robotics, industries huge and not so entertainment. Obviously we are looking into all these applications and these vertical ideas are not only for tracking but also controllers for robots. I don’t know, so many different elements that were horizontally going along our targets and some things were missing. So we virtualized those ideas and we brought them to startup levels.
We made joint labs. I have had 3 or 4 really big joint labs with big companies, to try to take some of them to high TRL technology readiness levels, levels to make sure that they are being absorbed and utilized by industry. Because in our team, I think we love doing basic research, which is the spirit that you really, you know, pushes the boundaries.
But at some point you really at night you question yourself how useful the stuff I’m doing. If you have a balance that some of the stuff that you’ve been doing are being used or being potentially studied to be used in industry, in health care and everything that makes you sleep better if you only do stuff that without knowing you really know where it is going.
I feel like it’s a little bit of waste, but of course this should be balanced without really exploring. You don’t know how to exploit it. So this is a very, very important trade off you need to make in your career.
Q2: Thank you. That was a wonderful explanation. I’m curious, you mentioned that you have several industry partners. Can you say a little bit more about what particular industry partners you’ve been working with? What are the kind of particular, you know, workplace situations that you’re designing for in their specificity?
Arash Ajoudani: Right. So, my collaboration with industry came in three different levels. Part of them are through large European projects. So large European projects, usually you have industrial partners that you design to technology through a co-design approach with them. They come with the problem, you fit the technology you build with them. And you do.
You do a lot of collaborations, for instance, in a project, your big European project that was coordinating, we had Volkswagen, we had Hedra, a big equipment manufacturing company, machinery manufacturing company in Slovenia, and also small SMEs in the Netherlands. We went and interviewed more than 1000 people, 2000 workers before, during and after creating a technology.
So we said, how do you avoid what you like the most. And then in the meantime, we showed them, is it what you thought of? And at the end we said, okay, now let’s really bring the KPIs on the table and see how they evolved. So this was like more of trying to understand how we can fix their problems with the technologies that we have.
It was really important for us because there were some partners, like Diane the German is to personalization. We even introduced, and contributed to standardization in some of the projects. So we introduced a new instrumental tool to understand how the how actually the, the, the sensory information that we can get from, from the workers to interpret their performances, their safety, their, stuff.
And then we made Incentivization the second way we worked with industry where we were really vertical projects. So the industry came to us and said, look, we have really this, this, this problem, can you fix that? And then we work with them. So there was without mentioning names, there was a company they really wanted to work for, there was a manual, part of the work that workers were spending so much time and effort on.
It’s such a simple task that needs a lot of precision. And the people that could do this were few. So they need to really focus and perform. But for robotics, it’s really easy because precision for robotics comes really for free, right? You just need to program in the right way and you have it for free. So we made it fully automatic.
And some more examples of this type. The third one is through a joint lab. So joint lab means a really big manufacturing company, let’s say like Leonardo came to us and said, look, we have a lot of requirements, a lot of needs here and there, and you have answers to many of them. It’s not one project, it’s many of them.
Let’s create this joint lab. We bring our people, you bring your people, we create this knowhow, and then we contribute to making the systems more autonomous. The cycle times are smaller, the performance higher, and the health of workers better. So the idea was not to replace workers, especially in cognitively stimulating tasks, but to help them to do their jobs in the best way.
But we were absolutely against putting people in, doing, tiring, simple, really repetitive tasks. And then we try to automatism as soon as possible. You know, always people say what should be automated but can be automated, should be automated. But I think it’s a little bit more complicated than that. Right? So we really need to, to look into human factors better.
Yes. Automation always needs to be there. But we need to be really careful about throwing automation everywhere because most of the time the adaptation needs the requirements of, you know, mass customization and things like that really makes it difficult to automate. And the human component is really essential there. But the most difficult question to answer is, especially when it comes to industry, what components are robust enough to bring in terms of robotics, in terms of human, you know, modeling and stuff like that?
That doesn’t really make workers feel frustrated. That doesn’t really make the company owners feel depressed because they invested so much money. But the system doesn’t really work. Every now and then they have to fix it and stuff like that. There are so many components that before we introduce the technology and raise their expectations and to make sure that this is working—
Q3: And maybe to take you back a little bit, you mentioned where you came from, more about the bio engineering side. Right. And then you went into providing. So where did you get introduced to robotics initially? What was the moment that you thought, this is kind of a productive way, right?
Arash Ajoudani: Right. So, during my math master’s thesis, I was working on functional electrical stimulation. Functional electrical stimulation is where you stimulate muscles, and then you create tension on the muscles, like the way we do. And you create movements. So if you do it controlled, you can create controlled movements, like walking, like reaching and grasping and stuff like that.
It’s particularly important for people like paraplegics because they don’t have movements. And during my masters we used functional electrical stimulation. So there are some sensors actually simulating pads, simulating pads you put on the muscles of the legs. We put them here, here, here, here. And major muscles. And then through controlled stimulation, stimulation, we made them walk.
So they had walkers, people that have no absolute control feeling or motor behavior below their waist. We made them walk. It was a long time ago. It was like 25. So I published several journal papers on that and during my masters, but it was amazing. So it was like, in a way, I was controlling something to move without the system being able to move, and it was so satisfying for many reasons.
People that were walking with this system told me I was telling them, how do you feel when you walk, even if it’s 1015 steps? You said, how would you feel if you could fly for ten 15 seconds? So for them that cannot walk for years, being independent and using their own muscles to walk was so amazing and they made me feel like I’m doing something for them.
But on the side, it also helped him a lot because when you’re walking, the blood flows more. When the blood flows, you don’t have any complications from sitting so much time. The muscles don’t shrink anymore because you’re putting so much. So it made me feel really good. But at the same time, I wasn’t really sure if I am strong enough inside to deal with patients because I was getting emotional when people were telling me anyway.
And every time, like, there was a guy that was so young, he was jumping at the sea, he got spinal cord injury and so young he couldn’t walk. So these things for me were a little bit too much to deal with. And I didn’t want to have a career on that. And I always see patients and I would, you know, if you’re a doctor, you need to be able to not absorb.
But I couldn’t do that. And I said, look, I love what I’m doing, but I want to control the stuff so that I don’t feel bad for them. I thought about it. Robots. So actually I said, I’m going to bring everything I learned from biomedical engineering and control. I’m going to control systems so that I don’t feel bad for them if something happens.
It was robotics. So I actually brought human understanding human models that we use for stimulation. Now, these modules we are using for understanding and what makes understanding human states and all that. And then we said, okay, let’s go. And I started my PhD on that and I’m quite proud of it.
Q4: Where did you hear about, like what was your initial perception and where did you hear about robots? Did you see them on TV or were they existing in other labs in your university right now?
Arash Ajoudani: Robotics has been part of me since the very beginning. I studied my bachelor in Polytechnic of Tehran, and then I started doing a lot of projects in robotics. So we entered the competitions of robots for navigating. I’m talking about like in 2000 and early 2000, working on robots, navigating on their own and, and, you know, following paths and stuff like that.
So robotics was always part of me, but I always wanted to also do something meaningful. Now robotics has, you know, is everywhere. But back then, robotics was not really known. I mean, you didn’t really know what the potential of robotics was. Industrial robots picking a place and doing it a million times a day. But now we know robots can do more than that.
But before I wasn’t really sure what’s what’s you know, what’s the potential of robots because they were just robots made for simple, repetitive tasks. But then we started looking into that. And then I wanted to say, but I want to do impactful things, so biomedical engineering maybe will help more people. But at the same time, robotics started to gain momentum.
And I knew I loved robotics because I always wanted to control things. And it feels good when you can control things without, you know, without people getting offended. So it lifted. It was a really good moment and they said, okay, I need to cross. Join, my intentions, my desires, my knowledge, my knowledge. So let’s see when they merge and robotics was this point.
So I think one really important factor to make every research successful is its interdisciplinary nature. I know sometimes, for instance, you really want serious robotics people to find vertical solutions and move forward. And this is what change is about. That problem. But if you take a few steps higher, if you have an inter, if you have interdisciplinary vision, view of how things should be connected, also there you can make a huge, huge impact, especially when robots and humans come together.
If you’re focused on either side, it’s not going to work. If you are too focused on humans, things will move slow, inefficiently. It will be frustrating. People will not. If you only focus on the robot like, you know, only productivity, fast movements and people will never use it. So you need to really have an interdisciplinary background. And if you come a little bit lower in terms of control, understanding perception.
So this again started to get balanced. I was there. So I said why not.
Q5: Do you collaborate in your work with other folks from other disciplines? You talked a little bit about…
Arash Ajoudani: Of course. Yes. So what I love about the European projects is that, by nature you have to include a lot of to a lot of groups from different countries, different backgrounds. Most of the European projects I had the honor of working with, a lot of colleagues and friends from pure, let’s say, human modeling perspective, biomechanics, psychologists, pure robotics people, mechanics, design, mechatronics design of, you know, really complicated hardware systems, perception systems, you know, like how to perceive the environment, mathematical people that could build new machine learning, solutions.
So these are all types of, let’s say types of, knowledge that we need in the workforce. And we brought all this knowledge through different, different collaborations. But even without a project, sometimes you have a problem and you don’t really know how to tackle all this. You need some initial light on where, in what direction you need to go.
And I had the really I mean, let’s say the pleasure of, you know, especially in conferences like IRAs, humanoids to to find those people, to tell them, look, should we do something together? You have meetings, you start talking to them. You don’t really also need big projects to do that. And a lot of people are so willing to collaborate on different, different topics without even being funded yet. But a lot of those ideas, like ideas over lunch, became a publication, then became a project, or maybe became a startup, or maybe became like, idea of a huge European project or international project and stuff like that. So yeah, I think this international collaborations are key to success because you are what you are. You can really expand, you know, and, and nowadays, obviously with, with large language models, you have access to a lot of information, but you need still the human touch to tell you, look, I tried this whether or not this keeps you month and, we need those experts to say, look.
Yeah, we’ve been working on it. Let’s do that. Let’s try this way and that way. And yeah.
Q6: You mentioned learn language models, which, reminded you of, something that was also curious, though, which is in terms of your, research that there have been particular tools or innovations from other sources, not just your own love that have helped you can push your work forward or that it’s giving you directions or capabilities that you think were important.
Arash Ajoudani: So nowadays, yes, nowadays you’re using a lot of large language models. They can help you clean up the ideas again. They can save your time. Right. So, in the past you had to try and error, you know, just you will try and see what works better. But nowadays, through access to unlimited information, you can actually filter a lot of bad decisions out because you can say, I’m going to do this controller for the robot, I’m going to do this application, I’m going to use this model for fracking.
And then you start typing and chatting with these big, long, large language models. This is a good idea. Yes. No, they are not going to make the decision for you, but they can help you, you know, save time. And we use it a lot in the lab. Actually, you know, the thing is, I find some people saying, oh, I don’t want to use it.
I’m against it. But these are tools. I mean, it will be the same if you go to an industry. Say, I’m not going to use a wrench. I’m not going to use a hammer, I’m going to use my arms, you know, so you would say, but look, these are tools that make you work better, more efficiently. You will save a lot of your health.
And here they are saving a lot of time and mental health to do things in the right way. I read somewhere that, whatever is known to us humans is written, it’s probably available on the internet. So if these models already trained on that data. So you have access to whatever we know.
And this is amazing. It’s like some sort of collective intelligence on your screen. So you are learning from maybe a novel, you know, a medal winner somewhere and also somewhere that someone has been using some technology as a worker in some country that you have never been to. So you have access to everything through one interface. So by one question, you can just fuse all this intelligence and bring it as a sentence in front of you, and this will save you months.
We definitely use that. And definitely. But the important thing is not to take it as if words of, I don’t know God or something. You know, you have to really focus and see what’s happening. I was reading something really funny. Someone was driving, he was on top of a bridge, and some navigator told this person to move forward, and the guy fell inside the river.
So if the navigator tells you, move forward. And you said there’s a river in front of me, I shouldn’t move forward. You should stop so large language models can harness it. You should not really give them the entire credit, but they can help you, you know, just go with them until the end. But if they say something, you don’t feel comfortable, you have to stop there.
But you know, so you need to find the right way of using the tools that they have and other tools.
Q7: So over time that it will do, I don’t know, sometimes people and then the bias kind of sometimes people have found connect to the kind of, really changing a technology that’s changed your abilities, for example, and make a more real world a try or write about rosters like, and Turning Point. Have there been any technologies in your work or…
Arash Ajoudani: Absolutely. And nowadays, much more so. Ross was a game changer in robotics, right before you used you had to write your own connections, your UDP connections, and you had to write different things. If you want to especially connect so many components together, you have to write one large code, knowing about how things will be connected. But now with Ross, you can connect things easier.
You can make everything look modular, look modular. You can, you know, easily get those or the way coding is done now, right? Coding, you know, on different platforms that are available. GitHub made you really, really coding efficient GitHub, GitLab, all these different tools tools are excellent now. So the tools are really, really today’s tools are making robotics or AI engineers jobs really easy.
So you don’t really have to worry about the small things. Maybe 20 years ago you had to know how to define an array, how this coding is done and where the bug is coming from. You know, you would spend days debugging the code, but now maybe just a simple copy paste in one of these, large language models, or chatting through GitHub or looking through a GitHub or even the availability of the source code.
So you’re going there. I said, how do I detect someone’s hand with a camera on GitHub. It gives you a thousand solutions. Right. And, and this goes for a lot of, you know, other tools like hardware, a lot of modular hardware has been designed by people and open sourced a lot of hardware. You can just print them on your own for the cheap, you know, a little bit of money and build your own robot.
I mean, the world’s most beautiful and impressive machine learning models run on 100 or €200. Robots. A few days ago, I was talking to two of my postdocs. I said, guys, we should try this. We were talking about an algorithm. It’s a visual language model. Said, use a camera, and we do that.
And I said, let’s use this robot and that robot. And week after day came to me and said, we did that, and it works. I said, I never saw you in the lab. I said, no, no, no, no, we bought this 3D printable robot, this very small robot. It cost $100 for each robot. And they bought two of them.
It’s a small robot, like it is made of plastic. And they bought a small camera like, €20 camera, and they put it on top at home after dinner. They use the code, they train those robots. And they did exactly what it was supposed to do. So 20 years ago, this was a dream. How could you do that with €300?
€250? They built a dual alarm system with cameras, and they ran a very sophisticated AI algorithm on top and did exactly what we had in mind using Ros, using available source code. So this is the future of robotics, right? This is the future of how things can be easily done and tested before you really have to go and make sure that you have the right tools.
You have robots accessible, and there are so many people working on robots. Get your timing available. And at this end, I think these tools, hardware, software and others, are all becoming available, making our lives much easier.
Q8: So what do you think are some of the challenges and some of the things that are coming in robotics in the next five, ten years?
Arash Ajoudani: I think the challenges probably are, as has always been, to be able to distinguish what’s a bubble, what’s the hype and what’s the reality. You would never know because you have to explore different directions. Nowadays there’s a lot of bubbles before, maybe less, but now we get here and there are a lot of distractions. I wouldn’t call them distractions.
Some of them are solutions, but you don’t really know how to distinguish distractions from solutions because there are so many things happening. The community is really growing and this algorithm is here. So should we invest time? You go and dead end, or maybe the other algorithm would bring you a lot of solutions, right? I think the biggest challenge everybody’s dealing with is to distinguish distractions from solutions.
And the second challenge we are dealing with is people really forgetting the foundations. So everybody’s trying to do some sort of end to end that I forget about all the lessons I learned about robotics, kinematics, dynamics, control, all this, that physics of the systems. I don’t know what robotics is, I don’t know what forces are. I don’t know what kinematics are.
I look at camera images and I do this action, I connect things, and there’s a magical box in the middle that does for me. I think this is a big challenge. This is a good solution, but it challenges some of the students. I see they’re growing, not knowing what forces are. If you ask them, for instance, are you looking at the dynamics of the system and they have no clue how the physical system is made or what is a mass spring damper system.
These are the foundations, and these are becoming less and less important. And people are looking more and more into coding and putting things together and making things work. And this is a big risk for robotics. This is a solution because coding and making things work much easier. But we should not really forget about knowing the foundations. If you don’t know the foundations, you will make large mistakes and this will not create good, efficient, good and efficient solutions.
Q9: What are some of the most interesting questions that you feel are coming up for applications of the kind of work that you do that are on the horizon?
Arash Ajoudani: I think this is, this is a very difficult, question to to, to answer, but I think my personal opinion is, is, trying to, distinguish and try to understand how much machine learning, how much, how do you, actually merge the information that you have from the physics of the system, which was classical robotics with pure computer science, which is just probabilistic based, data based algorithms.
Each one of them has super nice strengths, and each one of them have their weaknesses. So at what point do we need to merge them? We need to collide them more towards data or more towards pure control. I think finding this is the biggest challenge, and everybody’s trying to find the solution. It’s not easy because, you know, either you need a lot of data or a lot of models and none of them for robotics is easy, you know, like text is one, the image is two dimensional.
You can easily get them. There are a lot of available, you know, you just Google mug and there are billions of mug images. But if you Google grasping forces how much there are a few datasets here and there, reliable or not, what kind of sensor they used and not many things are available on the control side. How many models do we have, we have models of this and that, but do we have a model of an elastic deformable rope?
Yes. No. What’s this? So this is this for robotics, especially when it comes to real applications. So robotics is not only about taking the mug, putting it there. At the moment most of the applications has been shown as the strength of mobile robots, humanoid robots to arm robots doing these nice things. But when you bring it into cutting, doing really complicated tasks that require physical contact, the situation can start to become a little bit more complicated, right?
Because you need forces, you need more data. And that’s the type of data you can’t really create so easily either. You spent thousands of hours doing calculations that are inhumane. I think. I don’t really appreciate some labs making people try things a million times. Thousands of hours of people doing that might be the only solution. But then they show a very successful task of the robot.
I said, look, it’s fully autonomous, but you have to also show how many thousands of hours a PhD student, a master’s student or someone you paid had to show that to a robot. So these are basically the difficulties, the challenges I think you’re facing with no one having the answer, but the community is working really hard to solve this problem.
Q10: Are there any kind of simulation experiments that can help with that? I know, I mean, physically, yes, but I’m interested to work on more human interactions. Is there any kind of work in that area that seems promising?
Arash Ajoudani: Sure. I mean, simulation environments are very, very, very strong tools. And, and especially for tasks that you can simulate, simulate the model. So simulations have physical models of the external world. Right. So when you’re throwing a ball then you’re simulating the gravity and it folds down. And so and simulator some simulations are really strong in simulating the physical world.
Right. That’s why, if you’re doing autonomous tasks like walking nowadays, why robots are walking. So all of a sudden, the ordinary robot is walking without a crane and with a really nice natural posture, because we put these robots in simulations and we created a lot of different terrains, and we train them and train them and thousands of hours without really, you know, we slept and the simulation was running.
Right. So this is amazing. But how can you simulate humans in these simulations and how can you simulate data, stress, their physical ergonomics, their movements? That’s the part that we really can’t. So you need to create a model of the human. And then the physics of the ground is easy to implement. The physics of a ball which is jumping up and down is easy.
But for complex systems, complex friction systems, mechanical systems, complex deformable systems, complex humans of behavior, these are difficult to simulate because these are all models. If we had models, we need to know how to control them. So it’s a chicken egg problem, but they’re simulations. This is the limitation of the simulation. So one person, some people, takes some data from the real world.
You bring it in the simulation train as much as you can. So you have some really good understanding of the initial kick off point. And then you bring it to the real world and then train on real data. This is this back and forth SIM to real real sim. It’s helping us. So there are so many different tools we are trying, but still everything is very exploratory.
Q11: One more thing I was curious about. You mentioned that you were talking about kind of moving things into startups, that you could turn some of your work into, a startup that you worked on that or the…
Arash Ajoudani: Right. Yeah. But some colleagues are turning every idea into a startup. I’m not really a big fan of that, and I’m not really waiting for the idea of my life to make it a startup. But I’m somehow a bit in the middle, so the ideas really have full potential. We are really trying to work a lot on that.
So the real move idea and another, it’s quite for the moment it’s confidential. But two ideas where we are pushing really forward, to make a startup out of that. One is on the assistive robotics side, one is on the tracking side. So Real Move is a multi-camera tracker. And we know that this is potentially this will help a lot of people and in different cross domains.
And this is a startup idea we are working on. And, another thing is helping people to walk. So it’s a robotic system that makes people walk. So these two ideas are you’re taking them in parallel hoping that, you know, right impact right amount of, you know, energy and spend and and and things.
Well things will go smoothly but nevertheless technology is part of the solution. So we need investors and it really depends where you are. I know in some countries like China and the US, investors’ approach towards technology is easier, easier in the sense that they are more available and more risk aware and risk accepting. But in Europe, the situation is not like that.
Some people will come and say, oh, give me the best, give me the final solution, give me the solution that I know it’s going to work. And I would respond to them if I had the solution. I will go to a bank and I will just get a loan. I wouldn’t look for investors. So the difficulty is to find investors that they want a risk, but they want high gain.
So it’s a high risk, high gain situation. And some countries understood this better but some countries did not. So some investors do not want to risk, but they want to gain from the beginning. But this will not work like that. So we are trying to find the right people with the right mentality to, to invest in some of these, the right amount of investment.
Because if you want to get good, if you want to create a startup with a small amount of money, it’s a high risk for us, you know, you can maintain the quality of the work. So that’s the big struggle that us scientific people have, because we don’t have a team of scouting people that they go find, right, investors and stuff like that.
So we are scientific people in our own world. We try to try it with our network. We try to find investors, but hopefully this will just find its own way.
Q12: So anything else you want to share about your work and when you left out?
Arash Ajoudani: I think pretty much we covered everything.
Q13: Okay then we can ask you a final question, which we ask everyone, which is if you are talking to students or people who are interested in getting into robotics, what would you recommend or what would your advice be?
Arash Ajoudani: This is something I always also tell my students, it doesn’t really matter. Sometimes when you go to conferences, you see everybody is doing something and it creates a lot of interest for you that I want to do that too. But if you do that, that thousands of other people are doing the way thousands of other people are doing, you’re just becoming one of them.
So it’s hard to stand out.
It’s good to be inspired by what other people are doing, but if you want to stand out, you need to really add your own thinking. And you know, imagination and, you know, out of the box box thinking to that to be able to stand out. So some people. So instead of a follower follower, a complete follower approach, we need to start thinking about what can I do to have followers.
So what can I create that creates impact. So and in the beginning it’s really hard and it’s really demotivating because you sometimes you’re the only one in a session is speaking with three people in the room. But next door is the same thing that thousands of people are doing. It’s filled with people. So my only suggestion could be, don’t get demotivated by that push for if you think this is the way, just move on and one day you will see that your stuff will have the audience of thousands of thousands of people because you created the solution or hype or whatever, you created something important.
But if you become part of the trend, you are one of one of those thousands of people. Maybe you will have a big impact, but it’s harder in you’re not creating a solution. It’s just, you know, if your heart and mind are set and are telling you that this probably will go somewhere nice, even if people around you are not paying enough attention, this will, this will come, come across.
I give this example to my students saying, Kayleigh told several times that the earth is round and they put them in prison. So it takes really time to put really right and nice ideas on the table and people accept it. So we should we should not give up really quiet.