hello and welcome to the latest episode of the re automated podcast today we are joined by Alban dersy who is co-founder and ceoo of French company inbt specializing in using AI to locate moving Parts on Automotive manufacturing floors Alban is responsible for specification implementation and support of her company's product guide now which has seen great success in automotive companies by reducing the amount of fixturing and workholding needed for robots Alban thank you for joining thank you for the invitation so Alban how do you see the role of AI evolving in modern manufacturing environments particularly in the automotive industry well it's been a complete change from five years ago until today uh now customers are actually happy about having AI on their production line and promoting it internally and externally five years ago was completely different story but I think what really changed is that now people know what are the benefits spr by Ai and these are benefits that they really require which are more flexibility so more intelligence to the robot and then easier access to robot programming and these are two things that AI is making possible that also reduces the cost of automation so in the automotive industry that's ultimately what they're looking for fantastic and can you share some more specifics about how this flexibility and improvements and efficiency are coming through through the AI technologies that are now available yeah so there is two things there on one side there is during the programming and on the other side is during the runtime and during runtime the flexibility will come from computer vision like Vision sensors 2D cameras 3D cameras and we will have ai based algorithm that will analyze data from these Sensors how you use these sensors to improve the flexibility in manufacturing is to enable robots to work in what we call unconstrained environment so environment that is not typically made for machines who are traditionally used to do the same thing the same way every day but rather for environments that are made for people for workers and this is what we need to automate right now customers were used to build factories around production lines 30 years ago which this is the the pictures that we all have in mind the body and white lines and like 30 robots on each side that's not possible anymore now you can't build another Factory you have to do more with less and thus trying to automate what wasn't impossible before but now you can do it thanks to Ai and the intelligence that it provides to robots I think people our age probably remember Super Bowl commercials from like 2004 esque of the the Ford commercials the the large not a human Insight robots swooping in to a very structured environment so can you you said uh like unstructured environments can you explain a little bit we've we've heard this word of unstructured versus structured can you explain a little bit what that is and and why that's important for automotive companies in particular yeah so structured environment is when you know that the car is going to come at the same position the same way every time in front of the robot this is guaranteed thanks to jigs indexing systems fixtures tooling that are usually customade and just like mechanical mechanical Parts they're really maintaining and constraining the the robot and the part this constraining of the environment also comes with the cells you know the the cells and the around the robot that will protect it from people or from other things happening outside in the factory an unconstrained environment is the type of of environment in which we navigate every day it's an environment in which my cup that I have here can be either this way or this way on the table but I will still manage to pick it up because if you want to automate that cup picking process says in a constraint environment you would need to well have a jig to really hold it at the same position every time that jig will be custom made for this specific cup but in my factory I have 30 different cups I can't afford to do 30 different Jigs and change every day I need to work in the same unconstrained environment as a worker would do as I am currently doing so this is the kind of unconstrained we can also talk about unplanned environments which is an environment in which you don't you have some unplanned even EV coming in at the station or um that are not previously uh or human coming in next to the robot something that is also an unconstrained in in PL environment especially in automotive factories are moving assembly lines moving assembly lines are well the definition of Automotive manufacturing is the key for productivity these are the engines moving is the car is moving continuously these are this is an environment that is typically made to enhance human productivity but impossible to automate you have no idea how the car is going to be you can't track it a robot can't adapt to the position of the car and this is a type of environment that is by definition unplanned and unconstrained but that you need to automate if you really want to improve your productivity as an automotive OEM that's great and Alban can you give an example of an application that your product guide now with its AI powered technology has been successful in in the automotive environment yes so we develop as you said gy now which is a combination of hardware and software so it's based on a 3D camera that we Mount directly on a robot and an AI powered software that will analyze data from this camera at a very high frequency to identify the position and orientation of a part let's say an engine relatively to the robot and then realign the trajectory of the robot in Real Time according to how the part is moving what our solution enables right now is the automation of such moving assembly lines uh which are now possible to automate by just Wheeling a robot and a camera and let's say a tightening tool and perform a engine tightening process where it used to be not possible very good and and why have you chosen to use collaborative robots together with your AI powered Solutions we're talking about environments like production lines final assembly lines which are 100% manual don't be fooled the auto like the automotive industry is highly automated but when you go into a factory most of the people most of the production tooling you see are people and not robots so all the lines on which we work in on which we Implement our source solution to automate moving operations in moving are actually lines in which you have people on one side and the other side of the robot so if you want to be able to automate you have to be able you have to use a cobot that will be able to stop and go into a safety mode whenever there's a worker nearby this is really the key element I think about automating moving assembly lines very interesting so I I know from personal experience as well uh AI is sometimes seen as kind of unreliable I think many have experienced chat gpt's hallucinations where the model is like confidently incorrect so how do you ensure at inol that you have reliability with artificial intelligence in your solution um in automotive companies where you know they they need to be sure uh that things are going to work they have very low um tolerance for scrap and for failures yeah well uh we do have what we call a hybrid approach so on the one slide during the um so the way you program our system the only data that you need as an and user is a cad model of the part that we need to locate and track so the way that you would use our solution is by uploading this SC model into our software that will automatically train an algorithm on this scan model and this kind of automatic training this is what is powered by AI there is no input no fine-tuning no parameter tuning by the end user on the programming of this algorithm and then during the the runtime this is where we use a deterministic approach so during the runtime whenever the part comes into the field of view of the camera we can guarantee according to a certain what we call heat map that we can detect it a 100% reliability or if it's a bit further it's 99 or 95% reliability so we can actually guarantee in certain conditions 100% reliability of our product during runtime for the End customer very interesting I've heard that uh heat map I've I've read your website and some of your literature before can you explain a little bit more about what is a reliability heat map so that you can paint a bit of a picture for us yeah so whenever you have what we call a unconstrained environment you need to and whenever you have to locate a part you need to know whether or not you're well either locating it right or you're not locating at all right you need to have that kind of like safety data safety configuration that you send to the robot you can't send false positive all the time it's absolutely not possible and thinkable on production line what we have at in Bolt is according to how the part will enter the field of view of the robot according to its position and orientation we will detect it so which means we will match the 3D data from the point the point clouds from the 3D camera with the c model of the part with a 100% confidence or we will there are some times with like if it's coming a bit further away that in 5% of the case we won't match it but in 95% of the cases we will match it it's just a matter of like having a threshold in which we know that we can with certainty locate the part or in some cases not located but it's always about having a yes or no it's not about sending fake data to the robot that's absolutely not possible so I imagine kind of a a 2d almost normally distributed where the center is kind of green and then the edges are maybe red and and if if you're fastening on something and and the car is coming by I imagine that if the car is coming by and it lines right up in the middle of the heat map in the green part then you're very confident that it will be good and then no it's not we're not very confident is we know we're going to locate it if the car is coming in like I'd say like 50 CMS away from this green square then in if it's always case like 100 times it comes 100 50 centimeters away then in 5% of the cases uh we won't locate it but this is also important to know what are the conditions in which you will locate it for sure and what are the conditions you won't locate it so that not only you say it's a white and black decision but also you know why and the reason why is because it was too far away that's it yeah interesting so it's also the Zed it's the the depth as well and that allows Manufacturing Engineers to design you know their specifications uh such right exactly and it's actually the same on the production line this our cases are actually rarely happen uh in most of the case whenever we program a system we always deliver a system that for customers will work at 100% reliability because in such Industries uh I mean you have certain types of big oems and customer success is the most important thing that you can ever think of very good so pivoting a bit um let's look a bit at the future so can you provide an example of a breakthrough or innovation in AI uh for robotics that you find particularly inspiring or impactful yeah I think that what we're saying for AI and Robotics are so Ai and Robotics is a bit tricky because something that AI uses a lot are data set right and for a few years ago we didn't have data set in robotics what yeah could have it what we're seeing right now is the emergence of this data set that are being collected and like created over the years and I think covariant is a very good example of that having created this huge data set from all the data from all the robots around the world to enable uh this kind of like easy offthe shelf solution for autonomous robots for pick and place in certain industries that's a great thing that we're definitely looking and I think has a great potential in some types of Industries like Warehouse or e-commerce fulfillment this is huge potential um on the other side in terms of like AI what we're seeing is the type of like technology breaks uh are becoming increasingly accessible to the end customers who will be able to not only have great performance out of it but also learn how to use it it's not only AI now it's not only like in the dark carner for R&D it's not in a product stage like how can you implement it within your existing product using the foundation models that other companies like Nvidia for instance are creating yeah very interesting so there's more you know proprietary data that's really coming to light to be able to train these Foundation models and at the same time AI is becoming more accessible to more people to be able to use exactly very interesting yeah I've read about nvidia's Foundation model they call it Groot for example it's based on humanoids and I I agree that I think some of the challenges may be getting enough data for it to compare it to something like GPT you know GPT 4.0 is trained on billions of parameters they say kind of the size of the internet you know uh so it's going to take a lot of data to be able to get to that same level in robotics yes but we're still at the very beginning but we're seeing like great growth and a lot of like momentum around it not only like the tech people tech companies robotics companies are really like on the lookout but also the end customers are super enthusiastic and that's incredible from what we're seeing I think indeed like nvidia's group model is is very interesting and humanoids have definitely a great potential in um household for instance for services or for just like going from point A for to point B where it might be easier to have access to this kind of data to really make the robots move around for navigation uh rather than just for complex TX tasks that are usually jobs like welding or tightening for which for now it's still really complicated to have dat access to yeah more more in the home than in Industry even yeah so is there a subset um that you're particularly excited about I'll give you the analogy so in GPT it empowers so many different things they talk about law they talk about health care they talk about all these different things that large language models might be able to help with in the future for robotics with the possibility of a foundation model there is there a particular subset of things that you're excited about what I'm really excited about is how to make programming easier and I think this is something that is one of the easiest and like shortterm application of foundation models or llms is really how to make because you know robots have all different programming languages they all like this different approaches to like how to control joints and path planning and having this kind of like more uniform but also simpler AR architecture to use as an end user would be great to really boost the adoption of Robotics in the world because even if we we see all these big names in robotics and the huge growth in robot adoption we're still at the very beginning there are new new companies new created every day new needs identified every day it's it's Unthinkable what's happening right now yeah we talk a lot at Universal robots with our new partnership with Nvidia be about being cognitively collaborative and I think there's still a own bonus on people using robots to kind of put themselves in the shoes of the robot um even when AI is in the application you still kind of have to think in this linear logical thing that you know some people just don't think that way you know there's a lot of different ways that humans think and so I think it's really exciting in the future that there's the possibility of being able to be more cognitively collaborative and having the robots meet the humans where they are yes so than whenever logic is is involved this is will be game-changing for the future for sure yeah and then my my personal favorite thing that I'm really interested about is manipulation and grasping oh yeah because you know we I call it the last inch you know they talk about the last mile delivery in logistics a lot but the last inch I think in the contact between parts and robots and that interface is really exciting I I think that that will have some great advancements with some of these Foundation models yes that's definitely that that that is the key and there is so many things behind that it's not only like the robot and the robot control not only the grass Point identification but it's also the end factor and like control of everything simulation of everything so when we talk about robotics sometimes people stop at the Sixaxis robots it's not it's so much more and this is what you're showcasing Ator robot with the ecosystem that you've kind of like understood rather fast is that it's not only a sixxis robot it's so it's everything is intertwined whenever you go to a customer you never go by yourself you go with a robot manufacturer you go with the integrator you go with a vision system provider and all the small village is making something happen that will work 100% of the time yeah beautifully put yeah so any final words on future Trends towards the idea of how Industries can prepare for this rapid advancement in Ai and Robotics in order to stay competitive well the how the and I will talk just about the manufacturing industry because this is really my my my my basic and my bigest understanding of the market right now is the manufacturing industry has a ton of data let's say they all have the 3D models of their parts and all these 3D models of the parts they are manufacturing are great basis for having synthetic training about it's just a synthetic data and you can train algorithm on that data on how to better locate a part or how to detect a grass point or how to do a lot of different things so they have to make sure that they can get access to that data but also capitalize it and use it for people providing like US Vision soft Vision systems for uh robot guidance but also in data ALB exactly what do you mean do you mean pictures or models we only need 3D models and this data already exist it's an incredible thing that the industry the manufacturing industry has access to which is the digital continuity from design to manufacturing and we're currently seeing the gap in between the two from this conception to production narrowing down because we are using in the production the exact same data which are CAD models 3D models of what they using for conception very interesting very great thank you I'd like to spend our last few minutes together talking about you as an individual because I think you've recently won an award about being a a strong leader and woman in technology um so I'd like you to please share some of your personal experiences and insights as a woman who is navigating the field of Robotics and artificial intelligence yeah so I uh launched in Bolt with two co-founders a few years ago and we do indeed have a very deep Tech products uh in AI Robotics and most of it in the manufacturing industry I travel travel all the time to every country in Europe and the the in North America to visit factories to meet with customers and it is true that we are in a in an industry and in an area where we have more men than women that's a fact and there is many day in my life where I don't see a lot of women um but this is something that is also changing a lot um something that I've been able to see is that the automotive industry is something is very International so every company is has like factories all over the world so they had no this incredibly diverse culture in each of these oems and I'm seeing an increasingly n big number of people of women heading factories heading production sites heading teams and discussing with me and collaborating together on transforming how manufacturing is done today fantastic and and what do you think are some of the key factors to help increase the representation of woman in in the field is there is there something there well of course uh it comes so we have different scale of making that possible uh what we're really focusing at on inol is having that representation within our own company which is super super important so we always try to look for the women because there's always a gap in between the number of applicants between men and women so we have to look for the women extensively uh and to to be able to at least interview them that's super super important also I think visibility is key and there is an incredible number of events opportunities uh Awards as I just recently had received to really promote women on the stage uh put the spotline on them and give them a I would say a mic like today to express their ideas not only about what it is being a woman but about their business about what they do like the the company they're building which I think is is a great and they're also their in sights of the on the industry which I think is a great way to definitely promote visibility for women and then yeah I mean the the it's also about you know fostering stem among young girls it's stem education is so important I'm not an Eng my sister is an art teacher and she talks about steam now which is the inclusion of the a in stem so trying to bring in kids who are or children who are traditionally a little bit more artistic there are ways to bridge that Gap to to stem as well I think you know cading you know cat making CAD models is like a fantast or digital digital drawing and design in general typically it's a pretty right and left brain thing right so that's a good example Maybe as that's definely something that you see a lot in architecture as well like 3D design all that kind of thing right but I think a I'm not an engineer myself I have no background in like mathematics or physics or computer vision and but a lot of my customers think I'm an engineer is just because I started my career at in Bol by asking just so many questions and if you don't have any if you never ask questions you never get the answers and sometimes the answer is just so simple and the questions are also simple to ask um so I think that's also a great way to gain knowledge gain expertise and also realizing that even your customers or even the people you think have the answers sometimes don't have them maybe that's the uh the same answer that I'll ask to my final question for you but what advice would you give to young women who are interested in pursuing a career in robotics and AI well it's Pro yes it is going to be the same answer it is going to be asking questions uh everything is is simpler than you think it is in the end it's just about making a product that answers the customer needs so if you go from there which our customer needs we all have common sense to really understand them and it's then going backwards to build the best product possible but you really have to start from the customer itself and the needs are usually very very simple fantastic ALB and I could listen to you talk all day I think you are one of the most cogent guests we've had who have been able to come in and explain Ai and practical terms uh so I'm really excited to see uh this podcast episode get launched uh thank you very much for joining us thank you thank you for having me and once again this was the re automated podcast Alban is the co-founder and CEO of French company in Bolt who is specializing in using AI to locate moving Parts on Automotive manufacturing floors we'll see you again soon