Using AI in Robotics

Key Takeaways

hello everyone and thank you for joining today's webinar my name is Ree and I'll be your moderator today we're going to kick off today's session in a couple of minutes we're just waiting so everyone has a chance to join in the meanwhile though we'd love to hear from you so let us know where you're joining from using the chat or the comments uh depending on what platform you're watching on and yeah tell us something that you'd like to get out of today's webinar if you have any questions at all during the session today then please let us know using the chat and the comments and we'll be saving them for the Q&A as well uh one thing to note uh the session is being recorded and the recording will be emailed to everyone that's registered for the event uh you can register by heading over to data cup.com webinars you can uh scan the QR code on screen and you can also click the link that I've sent in the chat as well um keep your eye in the chat throughout the session I'm going to be sharing links to relevant bits and resources that will be uh yeah relevant for you if you're watching the session brilliant I'll be back to repeat these messages for any new join us shortly but until then enjoy the background music hello everyone and thank you for joining today's webinar my name is reys and I'll be moderator today we're going to kick off today's session in about a minute or so we're just waiting so everyone has a chance to join in the meanwhile though we'd love to hear from you so let us know where you're joining from using the chat the comments and yeah tell us something that you'd like to get out today's webinar um few bits of housekeeping for anyone that's just joined the session is being recorded and the recording will be emailed to everyone that's registered for the event if you haven't registered already then you can scan the QR code on screen you can click the link that I've sent in the chat I'll also be sending it again very shortly and you can also head over to dat camp.com webinars and find today's session at the top of the page as well as all our future sessions as well uh we are going to be having a Q&A during today's session so if you have any questions at any point throughout the session then let us know in the comments and I'll be saving them for the Q&A uh also keep an eye out in the chat for any resources that is sent through um these will also be sent out to everyone that's registered as well but it's always good to get them uh in the time well I think that's everything for me so now I'll hand you over to your host for today's session Richie Richie please take it away welcome to the webinar this is Richie now in the last few weeks I've had a few AI experts complain to me that every time AI gets mentioned in the mainstream news they show pictures of robots and actually they aren't the same thing because AI is all about algorithmically making predictions and decisions whereas robots are physical objects the thing is though while I get the criticism Ai and Robotics actually go really well together I'm sure you've all seen the videos of the Boston Dynamics robot dancing for example but that's only the tip of the iceberg So today we're going to learn about how is AI used in robotics Our Guest is Franchesco gadeta he's the founder and chief engineer at amethy Technologies and he specializes in building production software for big data analytics in critical systems franchesco's previously worked as Chief data officer at an AI startup and as a data scientist at Johnson and Johnson and he's also the host of the data science at home podcast and on top of that we actually have a surprise second guest so Franchesco is joined by elizio fante uh he's a vice president at the Technology Innovation Institute and he's got 15 years of experience in robotics AI swarm research R&D and teaching so uh without further Ado uh please pleas take it away from Jesco thank you Richie thank you for uh inviting me here of course and uh it's a pleasure to uh share what I know about Ai and Robotics and probably give some advice to the followers and the audience mainly coming from uh data camp and or for data Camp uh so yes I am uh uh Franchesco and of course this presentation is co-authored by my G friend eliso there is a pretty long story between Elis and I due to the fact that we started uh long time ago now I think it's what 20 years uh same University Pol Technic Milano uh and then our career um actually diverged only geographically because we we both stayed in artificial intelligence and Robotics field um and eliso is of of course uh um faculty member at the University of Amsterdam uh while I took another path uh more on the commercial side of things uh building my company tics and then of course um U founding the podcast data science at home at datascience atom.com uh among other things um I guess Ed is in the audience so if you want to say hi I would like to thank you everyone for attending and Francesco for the nice introduction also Richie yeah I mean nothing much to add uh I have a training in AI as well as you but specialized in Swan robotics in the last 15 years all right so I guess we can start um it's just a bit of background what we do at Tic Technologies um which is what people Define as a boutique company operating in engineering uh and artificial intelligence and machine learning Solutions um most of the time we provide uh solutions for what we call critical systems uh which is which are systems in which many times real time is involved performance is involved uh code safety is involved uh so many of these characteristics in fact are um typical of um uh AI algorithms and definitely of the combination between Ai and Robotics uh due to the fact that and I will expand more in the in the next slides uh in robotics usually not always but usually there is something that moves uh around an environment in which there are humans uh and things can also get dangerous uh in the most recent past of my company we um also focused on other sectors um uh industrial Robotics and uh of course agricultureal technology uh and uh recently in defense solution solutions for defense and Military uh so in this presentation I would like to uh start from some definitions because uh there is a lot of misunder understanding these days when uh it comes to Ai and uh uh robotics for sure um especially when it comes to defining what AI is and what robotics is so it I would take some time to um give a brief explanation of what these things are and how they are combined somehow and how they interact with each other and then I will proceed with some success stories uh in Ai and Robotics uh I will mention also some of the techniques the typical AI techniques that I used uh when it comes to Robotics and of course you know it's not gonna be an exhaustive list of these techniques for sure we need a the literature is very uh rich and uh we would need several hours which we clearly don't have uh hopefully I'm gonna do a decent job at least identifying the uh L let's say the the the right um hints for people to go and search and expand on uh uh on on on these techniques of course so well yes uh let's get started and uh uh the very first thing I uh want to U discuss is what is AI uh essentially what do we talk about when we talk about AI uh there is a uh a huge misunderstanding of what AI is and what it can be or can do for us um especially when people start discussing um AGI but before getting there last Preamble I promise uh is a quote by probably my top two favorite scientists in the in the in the world my personal view uh more than 50 years ago you said something like that I usually I visualize a time when we will be to robots what dogs are to humans and then rooting for the machines this quote is older than 50 years and still very very modern given the times uh probably if Claude Shannon uh could say something similar today he would probably mention cats rather than dogs because cats not always do what you ask them to do and probably that's kind of the relation that we have with AI today uh not always the AI that we are using today does what we ask thei to do uh with this said let's move to the definition this is one of the most um General decision uh General definition of AI uh so we start from General AI also called artificial general intelligence uh many people are talking about this um and the misunderstandings here are um non- negligible to say the least so what is Agi uh general intelligence is a form of intelligence that uh is formed by complex traits uh and there are usually many complex traits that um let's say fabricate the concept of general intelligence um it's not just a combination of several traits it's also the selection of the most appropriate trait for the job or for the task that you want to solve so it's pretty much like you know what we humans do when we are facing a a task or we have to achieve a goal uh we first of all select based on many things experience the environment what people say feeling and many other things that we cannot really quantify because they are typical of the human aspect of things things like awarness fear passion conscience these are all let's say quote unquote things that it's it would be very hard to U represent in Silicon or with algorithms we are very far from there as you can understand uh it's relatively hard if I if I can say impossible uh to get there with the current technologies that we have or with the uh essentially linear algebra this is what we are doing when we have these massive neural networks down there uh what we usually work with and when we speak about artificial intelligence we usually refer to a form of specialized intelligence um which is usually a single trait of intelligence um in the most complex scenarios one can find two do very rarely three traits of intelligence in combination other than that uh it's CI uh and the here we are mentioning of course the typical learning tasks um playing a video game by watching how other players uh played uh Planning by watching how people plan tasks relatively simple tasks and then language and with language in the last few months or year year and a half we have seen this GPT family of model and the companies that we know uh claiming that they are close to artificial general intelligence we're very far from there I want to emphasize want to stress uh uh enough It's never enough in fact to stress around this those who know me from the podcast know how much effort I put in uh staying away from artificial general intelligence and call things the the uh the term they they they they have so my question probably to eliso is uh who's an academic in in some under certain con concerns is U why there are so many reputable people uh with very good reputation especially in Academia who even there they claim that we are close to artificial general intelligence that's something that I still do not understand look I believe it's about expectation is what people expect and this is this is because of the movies the the books like Richie was saying at the very beginning have you ever seen a sci-fi movie in which a guy is dealing with AI just doing a data analytics or just programming a robot that maybe crashes it's always about a robot that is exactly identical to a human and that's an AGI so movie is all about AGI but the academic field and the industry field is not about AI I do agree with you uh I don't think a is in reach so far I don't want to be more Brave than that uh I think it's also linked to another topic the topic of Singularity with AI a lot of people even very respectable and reable people in the field like you know key player in the field believe that Singularity is in reach in a few decades Singularity is by the way the ability of AI to create other AIS and therefore accelerating the progress of technology I also personally do not believe in Singularity however it is sort of a very motivating thing to believe I do understand when you believe into that uh because it it drives your personal research agenda your personal uh you know Endeavor towards that direction makes sense so staying ambitious is a kind of key in uh that's why people speak about hii that's what I how I get your answer thanks Alis thanks for your contribution so a bit of History um now uh AI of course was not born yesterday uh how did we get here um it it was not even born in the 80s as you can see from this slide I just use the time span of my personal life I was born in the 80s but in fact AI or attempts to um work on AI um started probably in the 50s and here again eliso uh will probably add something for sure later um so in the 80s we had what is is called expert systems uh pretty much if then else for those who are familiar with programming struct uh constructs uh there is a rule or a bunch of rules uh an expert uh who defines these rules according to the domain or the task that they want to solve and there is a computer that executes this rule if a condition applies the algorithm does something if it doesn't it does something else and so on many of the things that we have seen even of 2000 that we believed were AI uh they were actually rule-based systems the first chatbots to say something uh were actually rule based and then we had semantic analysis and other fancy things but before that uh when machines were not that powerful expert systems uh actually ruled pun intended um then we had uh machine learning Renaissance essentially uh the start of the internet for the big public the 990s um this was when we started seeing the first recommender system in actions because recommender systems come from the 70s actually so if you think that Netflix invented the recommender systems or Amazon or whatever book uh um um whatever book uh uh recommendation you get uh that stuff came from the uh from the 70s um it saw the light in the '90s uh and in the '90s we had like the internet uh so when people started actually producing um data and sharing data and uh storing data and and stuff like that but we didn't have enough data for you know to start calculating things uh and so you know data science How We Know It uh did not could not happen it could happen in 2000 U when we had indeed the rise of Big Data and deep learning we also had more powerful Machines of course um and uh in 2000 uh I want to stress on this because it was just 10 years uh and we had some really it was a flipping moment the moment in which we could do all the things that we could not do just 10 years before uh and 10 years in the lifespan in the time span of humanity is probably less than the blink of an eye uh so in those 10 years in fact we by producing data by generating data and also other improvements uh in the hardware for example um we could flip a lot of things um 2010 just 10 years later we started seeing the first AI based applications in everyday life U Google home Amazon Alexa even though they were not like pure AI the way that we know it now but uh they were definitely um you know supporting people in making decisions on everyday tasks and finally 2020 um AI for good and explainability uh ethics you know when we see AI becoming let's say more and more powerful we start coming with other questions that are like okay is this right is this ethical um is this Fair uh is it safe and so all these discussions started you know got started just very recently because of the conditions in which we operate today with a eye and of course large language models we all know uh everybody's on large lar mods today and we do pretty much anything and we gonna do probably much more in the next uh in the next few months or years um Elisa is there anything that uh you am I missing in this slide probably something like winter or season Yeah in our in our LIF SP we are both from the 80s we are not not missing anything but to put it into perspective AI is born in the 50s and from the 50s to the 80s I believe it went through yeah the what you mentioned the two winters of AI there were moments in which uh let's say expectations were not met from a business perspective and let's say funding agency stopped believing in AI stopped there stopped to be a a momentum in funding those projects I believe now we are in third winter but we should uh somehow listen to what our mentors that lived through those time told us that is yeah one of the things they told me is not to believe too much in AGI and I think we come from the same mentors that's why we have the same all right uh so some sectors um uh Ai and Robotics um are more and more present in many different sectors of course the first one we have seen um robotics especially um is manufacturing and Industrial so we know industrial robotics is where even today there is not so much AI there is much more Robotics and automation uh this is not an exhaustive list of course but it's a pretty complete one that um continues later um healthc care um is another sector that has seen a lot of Automation and data analytics uh AI not much used yet because of reliability issues um and risk of you know if something goes wrong and the health of an individual is at stake you know things can get um quite dangerous and also the responsibility is quite difficult to understand who's responsible if the coder or the doctor or the programmer or just linear algebra or or statistics um so until we Define uh these things or we Define these things in a I think legal framework it's going to be hard to see I in AC in critical sectors like healthare um agriculture is another very important sector to be honest with you is my favorite sector I I I love it so much when it comes to uh looking at robotic Solutions and uh also the combination of AI and Robotics in my opinion is much more present in the agriculture uh sector than in all other sectors probably uh space um is is another one uh um I will I will spend a few words on the agrix sector later in the in the presentation uh Logistics um very similar tasks um in which AI uh is um used or sometimes not even AI simply uh algorithms that provide vision and localization and planning and decision making the data way um uh defense and Military unfortunately we have seen especially due to the World conditions that we all know an increase of attention of uh um you know from the defense and Military sector to new technologies and in particular uh to the combination of AI and Robotics uh not only for the war zone but also for urban urban settings for example uh defense in the sense of surveillance um and uh of course the usual tasks here computer vision localization with some additional complexity because in the war zone uh GPS not always works for obvious reason and so you have to find other ways to to localize uh uh your target or localize yourself if you're a robot um there are other sectors of course education and entertainment um these are uh on the um on on a massive grow especially when it comes to video games U many workflows are changing in the way people produce video games today uh with the aid of AI and 3D modeling um and also Technologies like VR uh that's a very there is a good setting to CI uh in action in the next few years or less um then social robotics um assistive Technologies for you know people with disability or the elderly uh and last but not least space exploration uh which is probably one of the best I think it's one of your favorite Alo isn't it yes indeed it's my favorite because it's uh the most challenging one from my perspective you should get rid of all the assumptions on human infrastructure it's completely an infrastructures world the robots themselves plus Landers and orbiters of course are the only infrastructure you can rely on when we write grants through story We tend to take that application as the inspiring one so especially in small robotics sometime hard to find application Earth we have to look into space nice so you you said it what is what is robotics what is a robot so I guess that this is one of the most difficult question that I had to answer uh even to make this presentation uh because the answer is like it depends uh it depends who you ask to um so if you ask u a roboticist probably I will see it later uh we might have roboticist Among Us here um what is a robot usually they say it's a programmable machine designed uh to perform tasks autonomously not necessarily sometimes even just with the as minimal human intervention is possible uh there might be mechanical components with some sensors there might be a computer um and of course in terms of capabilities it's uh something let's say that can uh go from being a simple system performing repetitive tasks think about industrial robotics packaging or in logistics or just moving things from one side to the to another or more advanced machines that can perform complex tasks um for example planning for example solving problems more complex problems um interacting with the environment um recently even with humans human workers in in warehouses for example then if you ask someone else uh more you know close to the software world uh he or she will tell you well a robot is a combination of software uh that can be soft or embedded software U some Hardware some mechanics there are some joints there are materials there are geometries a humanoid is not a drone which is not a ground vehicle with wheels uh or a manipulator arm you know these are the geometries that uh might change uh there there is electronics of course there are sensors for perception there are there is a CPU multiple GPU dedicated boards uh and of course there's data I mean data is always there of course because it's it can be sensor data it can be events that come from the environment it can be input from a user um it can be many uh sources of data of course so we have data flows and a a software person will just say look a robot doesn't have no Hardware that's just a piece of software that can learn decide and control and if you think of GPT for example can be considered a a robot in that respect probably a bot what people how people Define B which is just pure software okay um I guess I should ask Aliso Al what do you think a robot is yeah franisco my favorite definition of Rob of a robot that I use in the classroom I I normally take the a picture from the book for by Richard saton andto the reinforcement learning and introduction for those that have study the I should know the book so you have an agent you have the environment and the environment gives feedback to the agent that can perceive the environment through sensors and then the agent through actuators can modify the environment so my definition of robot is you take that agent and whenever you give the agent a body a physical body with Hardware then it becomes a robot I see so yeah you have four definition V for me is not robot in my definition all right thanks then then we have a we have more definitions on I have to add another slide then all right good to know um I guess we all agree that robotics has one definition in fact which is a combination of disciplines among which I and listed this ones probably I missed something but essentially we have pretty much everything in in robotics it's one of the most interdisciplinary Fields or disciplines that you might think of we had pretty much the same we said pretty much the same years ago probably more than 10 when data science was let's say born uh in which we said it's a combination of between statistics computer science mathematics e economy and uh when it comes to robotics is way more interdisciplinary than that there's things that might even not match and they do like software engineering and mechatronic or perception and statistics and probability Game Theory uh telecommunication and uh control theory whenever you have controllers and motors and sensors and sorry and servos um that is data science um and and probably I missed something here said you want to add something or is it fine not really think T is quite complete all right so um the next question I try to answer is what do we expect from robotics um now there are some stereotypes here I have to be honest with you um and I apologize for that but um if you ask of course if you don't ask the the single Japanese will not mention assistive robotics but generally speaking there is a culture in countries um that is kind of driving them towards some type of Robotics and we know that for example Japanese they look at robotics robotics under uh behind the lens of assistive Technologies for the he people with disability Etc and actually you know culturally and historically they are kind of the best in the world for this type of uh of applications um in Europe for example we have uh uh of course the Netherlands uh where eliso is teaching uh is probably the best in the agriculture technology sector uh in Europe for sure but also in the world uh and you know the way they look at robotics what they expect from robotics is probably some application that comes from indeed the agricultural Spector um unfortunately uh you know all those countries that are involved in uh conflicts and Wars uh they probably expect um and actually they do expect that technology and Robotics U help them to um protect themselves or to offend uh an enemy um in any possible way uh and of course is uh um technology applied to uh the war field uh and I'm pretty sure that if we ask the average Italian what they expect from robotics they would probably uh expect something like that and as an Italian I actually agree with that type of Robotics and I still think that Italians are the best in the world for B like the Japanese um Ela what do you expect from robotics well more of those machines I guess I have to say the the best ones are done in Germany if I remember correctly but we just eat the the produce all right so some AI techniques in robotics as I said uh they're not it's not an exhaustive list um there definitely uh Locomotion which is uh also um referred to as path planning this is something that to be honest with you uh has been uh is being used or uh it's it's something that is very you know G video game developers are very familiar with this um usually the non playing character of a video game uh or even the playing character video game use this type of uh um of of algorithms for you know to move in a virtual world uh there is decision making uh people who are familiar with for example data science or the pure form of data science and the data analytics they know exactly what I'm talking about uh there is perception um object recognition via cameras um as well as liar or other or other sensors or a combination of these sensors that's called Sensor Fusion um in order to that detect something recognize something localize something can be a target a building a person uh or understanding a scene um we have seen this in computer vision examples when we had one of these bot software version of the robot that says um there is a kid eating an ice cream in this picture and a dog running in the park you know that's the scene understanding well that's actually seen description understanding is a different thing there is manipulation and interv vention these are usually techniques that are used whenever you want to let's say transfer uh some kind of feeling from the robot to the human um think about you know apics technology think about a surgeon who's operating a client a patient the other side of the world of course the robot is there the surgeon is in another continent or another country and he needs to feel I don't know the tissues I'm not a a doctor I'm not that type of doctor so but you know that there must be some kind of Return of these feelings from the robot to the human to the surgeon uh or even in gaming um eliso is also a very good gamer almost a professional one that I know of and he's very used to finding the very the most recent haptics interface isn't Ito yeah I didn't buy um of course there is also computer vision we are you know at a very Advanced stage uh when I mentioned computer vision because it's a type of technology that of course is applicable uh in many other fields not just robotics uh robotics can leverage computer vision uh for the tasks that the robot need to accomplish um task planning is probably one of the most fascinating fields if you ever want to expand on that u a huge amount of literature for task planning many algorithms many complications many challenges really something I personally advise uh for the students who want to uh touch that particular field and then there is eliso field collective intelligence this is also something that um I just want to give my very trivial definition and then I let you I don't want to step your feet but the way I see collective intelligence if I had to explain it to a fiveyear old is like uh maybe more than five older than five uh it's like sprinkling some you know a process in in many parts to multiple agents so that they can cooperate collaborate or even compete if necessary uh in order to achieve that task now tell me how wrong is my definition I actually like this definition because it uh drifts away the the view from the single to the collective which is this exactly how we should think anyway it's a field inspired by uh biology social insects in particular other social animals like birds and fish but also I like uh to take inspiration from physics that consider much much larger scale of collective systems so it's a very good definition the the key here is emergence emergence of intelligence emergence of perception emergence of different traits that's uh what we tend to do and also one of my uh research areas perhaps I wanted to mention another field maybe that could be listed here evolutionary robotic so evolutionary Computing for some people is AI for some other people is not AI if we consider it AI it could G race to a I mean it gives race to a completely new field a completely brand new field which is evolution robotics it's similar to machine learning in the sense that it can be used to tune and find the brain of the robots but at the same time it can also search the space of different things different other traits of the robots including its own bodies or the properties of the materials or the properties of the sensor and so on so it's a very promising field I mean not very new to be honest but there is a lot of recent development in that okay thanks so some success stories in AI uh I would say Ai and Robotics um I think that many of you have known have heard from Boston Dynamics um Richie mentioned at the beginning of this presentation uh they're successful in building robots but uh these robots are extremely expensive I don't see how the big public could actually purchase one anytime soon but they are amazing I mean they are definitely probably commercially less viable let's say than other um more accessible types of robots but if you have seen if you have watched these demos they are in incredible like Smooth movements Flying Kicks uh it's something that is really out of your mind um there is soft Bank robotics producing a robot that it's called now it's a humanoid U cartoonish robot to be honest used for educational purposes uh and then there is again my favorite U agriculture technology sector vertical farming um growing food in urban environments or where uh the weather doesn't actually allow you to have an open field um farmer farming uh and here we have aerofarms banic and infarm infarm is German um and also the usage of drones manipulators usually that's the combination of uh uh of you know the type of robots that operate in these uh in these fields um we have also prision agriculture there are some cool demos that I found uh on the internet I will show them in a minute uh F wise is kind of leader here uh and then of course multi-way and Locus robotics for uh robotic picking systems these are basically the robots that you see or you will never see in the warehouses so that package from Amazon that you received uh on time or before before time is probably due to the operation of one of those machines um and finally astrobotic uh this is we are on the more advanced um let's say side of things um space exploration and as AO mentioned before that's where things really get critical really get nasty but also where the biggest challenges are and uh you know we are very close to sci-fi when you operate in these fields um so as I promised there is some uh um uh some demos that I want to mention uh that are specific of the um agriculture Agri Tech sector uh in particular automatic collection um so as you can see here now these are demos but of course U you know these are pretty functional in the sense that it's not just a showof it's like something that actually people are using uh harvesting robots that work day and night and they can definitely uh operate at a very good Pace um then we have for example automatic or Precision spring this is a demo um of course probably we can get the render later but it's a very uh you know technology current technology is very close to uh getting there and uh uh there is yet another one uh which is U uh the combination of analytics and uh uh so data analytics and Robotics and AI uh in this type of context in order to increase the efficiency of certain operations you know typical operations of the agricultural technology sector where also data are used and data analytics engines are used uh in order to take decisions or to make decisions with the better support uh and finally autonomous operations so these are kind of you know uh where you see this is the scenario in which you see a a robot that actually uh is not remotely controlled completely autonomous uh you just you know throw it there and it goes now when I say this I have to be very cautious though because a person who sees this would think like yeah we have autonomous robotics we can do autonomous operations no pay attention to the fact that this environment look at the environment is very simple it's an open field straight lines no obstacles no kids crossing the street no grandma standing in the middle of the of the street waiting for the green light uh etc etc no dogs running Etc so there's zero complications uh zero CH not challenges but zero um uh uh destructions let's say Urban settings are not like that uh so don't expect that this level of you know autonomous decision making can be transferred overnight from the agriculture field to New York right um and please don't believe those who claim that run away from those people actually okay um which brings us to the common misconceptions uh of AI and Robotics and uh again I've been uh an advocate of keep your Fe keep your expectations realistic okay when it comes to artificial general intelligence please please don't believe the people who say that artificial general intelligence is here just laugh and run probably do both fast um the second meth is definitely uh claiming that autonomous decision making is possible as I said it would be possible but under certain constraint let's say simplified environments uh like the agriculture sector can offer moving to to more complex environment uh it's very very difficult um and the third myth is about artificial intelligence in robotics is just like artificial intelligence anywhere else this is another myth because things are different in robotics and probably eliso is better than me at explaining how different things are when we combine artificial intelligence and Robotics yes I agree so basically along my career I always seen robotics trying to catch up over ai ai running much faster and robot is catching up later and one of the reason is don't forget we are dealing with embedded systems embedded systems are you know smaller scale Hardware with more demanding power constraints uh we have seen only recently let's say advancement in embedded system in the last 10 years I would say 10 15 years compared to the equivalent advances in traditional computers that happened at least um maybe 10 years before so we are let's say uh there is a gap of 10 years between the two and uh I I think right now is a time in which we are seeing a lot of new uh development in terms of uh porting AI into robotics if you go to conferences in robotics I and AOS if you don't do learning you're sort of a loser nowadays it's very hard to even publish a paper if you don't have the word learning and so on and this is TT the fact that you have a very good hardware for example the Nvidia Jets on boards that are can be mounted finally on the robot and they can accelerate the the uh neural networks so it's all about machine learning that is running plus some end perception computer vision and some uh let's say perception into action type of uh approaches I wanted to mention the excellent work done both with my colleagues at tii but also in Switzerland by Professor David escaramuza drone racing drone racing is an excellent example there is an endtoend uh learning problem perception and control and the task is to per compete against human in racing with drones okay and it's true that it's uh indeed the the tooling is also you know it's very fragmented I mean when I got closer to the world of Robotics uh in the Years uh there is no standards so everybody's working in his own Silo kind of and then when things have to let's say be integrated the efforts that a you know a project leader has to go through is incredible and sometimes actually that's where you spend most of your time integrating things just because you have these very complex Stacks that have never spoken with each other and of course you have to find smart ways to uh to let them deal with each other um so I heard someone in the audience wants to work in in Ai and Robotics uh there are many things that you have to know of course as always uh of course it depends what you want to build um what you w to be uh where you want to be and how much time do you have of course how much of your life you want to dedicate um definitely if you have all of them you are in god mode speaking of video games uh but otherwise if you pick let's say two or three of these skills U you are positioned I think um very well uh with respect to to the rest for sure you need um programming skills I mean programming is uh the number one ingredient in everything you do with thei and Robotics um now there is a list of languages that you might be interested in my personal uh favorite of course is rust uh and there are many reasons why uh I consider rust the language of the next 50 years or more uh maybe if Rich is kind enough to invite me for another speech I will tell you why um there is mathematics and algorithms that of course um uh you have to master uh in particular line linear algebra calculus and probability these are usually uh you know the uh topics that you cannot skip um at school you better not um electronics and control systems when you want to deal with the actual robot uh so when you want to deal with a board that is controlling some Motors or a a compute board that has a dedicated Hardware uh with the dedicated sensors while then getting familiar with electronics and sensors of course uh is a must uh and then of course machine learning and AI but I guess that in this audience uh majority of the people who are listening or watching uh are probably very familiar with these topics even thank to the courses of of data Camp um and uh I can continue of course on more special um let's say software Stacks they call it Robot Operating System Ross uh in the in the community um I don't want to express my feelings about this but uh date but don't get married okay it's not really an operating system so even the name is quite misleading um maybe you can contact me and I will tell you what's going uh wrong with Ross or why I don't prefer Ross um with this said there are of course other let's say soft skills when it comes to um you know Ai and Robotics in combination for sure problem solving skills um and last but not least list uh continuous learning and adaptability because robotics is as I said one of the most interdisciplinary field that you ever have to deal with in your career uh way more interdisciplinary than what data science was 15 20 years ago so the the diversity of the topics uh that you need to master is pretty uh broad um I say am I missing some skills here yeah perhaps those at the more senior level like project management I find this project quite challenging to to manage when you are dealing with let's say six months timeline and you need to deliver something you need to know that even intermediate Milestone May Fail and when they fail there is a whole preparation that needs to be repeated to come up with the same Milestone which may de delay the project significantly and you need to know that in advance so those are the sort of soft skills you also need to to have if you go for more senior roads thank you so if you want to get started on a more practical side of things uh I personally recommend of course in case you want to also explore the magic of uh the rust programming language and at the same time work with robotics uh I think that opener they call it like that open rust robotics uh is a good start it's a don't expect anything super stable and super mature it it is not but in my opinion if you want to combine the two things like learning rust and uh well getting familiar with rust or maybe finding a u an open source project to contribute or you know just to uh improve your skills uh I think that that's a very good combination there is also very good uh um integration with Ross so in case you want to expand your knowledge with Ross and rust and Robotics uh these are uh this is a very a very good a good one of course it's pure programming um because if you want to do something like programming and building and definitely if you have a bit more let's say budget at your disposal uh you can definitely move towards one of these kits U that you see here in the pictures like Parallax uh and clear paath robotics is providing this the turtlebot I think it's version four now uh it's a pretty nice platform open uh so you can customize and let it play in your home uh pay attention to your pets they usually like to destroy these things and they are quite expensive um crazy fly is another it started as an academic project uh it's a very valuable one uh it's a swarm bundle of Nano drones that you can afford for for uh 4 5,000 something like that or less I'm telling you the uh you know the flagship product uh but otherwise we can buy even one I think it's a couple of hundred euros um and it's a a very nice platform to to play with uh there is also python involved so in case you don't want to you know go down to the low level uh definitely some Cera courses uh robotics flight um a very valuable one and uh of course Nvidia autonomous robotics provided by um Nvidia that does not always provide uh gpus for gaming but uh also very interesting open source projects in particular this one and uh several others um more in the in the field of simulation um Ed want to add something or all the big players I think are coming to robotics including Microsoft for example but but I would say that in the audience if they have a particular interest it's always better to customize the recommendation for depending on the particular robotics in which one wants to start and grow okay I guess we are uh at the end of this presentation I hope uh we did a decent job um uh tickling the interest definitely not exploring exhaustively highend robotics that would be impossible but in case you want to get in touch with us these are the emails and uh if you want to also work with my company and my team uh you can reach out at tic.com sometimes there are open positions and of course um if you don't follow um you can you know feel free to uh you don't need any subscription you just uh hit the browser or apple podcast teacher pbin or Spotify and uh you can get me from the podcast at dataset home.com thank you everyone and thanks foris s for the contribution all right that was fantastic both of you um yeah I I really enjoyed that um I love the quote from Claude shanon at the start I guess um of course like the anthropic clae um AI That's that's named after Claude Channon right so uh he he's uh he's he's pretty famous for the last few decades um so we're going to take some questions from the audience in the moment for anyone on the audience if you have questions either for Franchesco or for Elio then please do ask them in the chat now and before we get to that I've got a few questions so I think the thing that blew my mind there is just how many different skills you need in order to work in robotics like there was Elis like 12 different disciplines that all are kind of full-time jobs as they are so I guess first of all um where do people normally come into this field from like do you is it normal to start with a mathematics background or do you start with an engineering background what's the deal that should I take this one yeah since I'm the academic yeah so it's a very good question they they come from different backgrounds they could come from the computer science part which is where me and franisco come from but they can also come from Computer Engineering and me mechanical engineering aerospace engineering for the for the drones mathematics maybe not but you know you need to have some uh variation of mathematic more to towards Computing so maybe computational mathematic can be uh I think also physicist can learn other disciplines quite fast from my experience they adapt very quickly for for any job okay fantastic and um once you have like a whole team for robot assists like what what does that consist of you mentioned was inter disciplinary so what's a typical team like so this I can take because when it comes to team at least from a commercial perspective so a team that actually um beli givers a a product um to a client or a product in general uh asil mentioned it's extremely difficult to manage delays when it comes robotics um due to the fact that there is this you know restarting of operation so first of all you need people who can handle that on a personal basis like even with the soft skills being very patient of how the workflows go there they can look very chaotic um since the Inception of the project until the end so it's like the new way of producing things when it comes to the skills of course um uh they have to be very diverse in my personal case for at ametic uh we are uh much more let's say imbalanced towards the software engineers and so coders and developers but also Architects uh so these are the two major figures that you see in and Robotics projects uh due to the fact that there is a very good combination of design and implementation uh where design usually takes uh longer than implementation which is quite weird uh usually it's the opposite for Pure software projects that I've seen uh but in robotics it happens kind of The Other Extreme okay so it just seemed like um project management skills is crazy essential there as pretty broad sort of a skill set so that that seems useful um one more question for me before we go to audience question just is there anything that you're particularly excited about at the moment in the world of AI and Robotics like what What's um getting you uh out of bed in the morning at the moment well I mean one field that I mean I'm I'm following the progress of course like everyone else with the large language models CH GPT and so on there are some efforts to uh let's say bridge this with the robotics uh I've seen mostly in the field of manipulation I would like to see maybe contribute I don't know uh to other fields of Robotics and for example one of the things you could do is to uh write a specification for a controller or a perception module and have the larg language model generated of course it's prom to nonsense so you you need an ecosystem to validate this and somehow you have to advance also the research in larger language model to reduce the the nonsense but to me this is an very an exciting very exciting direction to go towards okay excellent and Franchesco from my side uh uh I'm not that much excited by large language models um due to the fact that um uh language is just one aspect of the world in fact of people um and you know the tasks that we usually deal in environments um not necessarily have to deal with people they can deal with many things so that's the first thing the second thing probably the the most the things that keeps me most excited is as I mentioned the agriculture technology sector uh due to the fact that there are many challenges that are uh kind of cross- sector challenges like uh uh robots diverse robots different geometries the weather can be bad there might be wind uh batteries can go uh can fail there can be water uh you know there is there are many environmental circumstances that make it challenging um and then of course what I personally would like to see in the near future is in this sector like mastering this sector I'm more a traditionalist when it comes to you know research or applied research it's like uh let's start from a sector let's conquer it and then see what we can transfer from there to all the other sectors like for example cities okay I'm going to let the two of you fight over whether or not large language models are a good idea after in robotics after this it's good to different good to have different opinions excellent um all right uh just before we go to audience questions as want to talk a little bit about upcoming webinars so tomorrow it's Thanksgiving we out stop so uh if you want to well for the Americans if you want to avoid your family dinner then please come to webinar for everyone else uh definit