Jimmy Carroll: [00:00:03] Hi everyone and welcome to this episode of the Manufacturing Matters podcast, where we discuss the trends and technology shaping the manufacturing industry today. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing, and I have the pleasure of being joined by S.K. Gupta, who is the co-founder and chief scientist at GrayMatter Robotics. S.K., thanks so much for taking the time and welcome.

S.K. Gupta: [00:00:27] Thank you, Jimmy, for inviting me. I’m very excited to participate in this discussion with Manufacturing Matters.

Jimmy Carroll: [00:00:35] Well, I appreciate that very much. So S.K., I’d like to start off by asking, for those who don’t know, could you maybe tell us a bit about GrayMatter, how you got started, and what you do there?

S.K. Gupta: [00:00:44] Sure. So GrayMatter Robotics is a company focused on delivering automation solutions to high-mix manufacturers. So we basically deliver people smart robotic cells, which then enable them to automate a lot of their manufacturing operations, such as sanding, polishing, buffing, grinding, blasting, spraying. So these are all fairly ergonomically challenging operations. So using our solutions, people can just bring a part to the cell, put it in front of it, and task it to go. And then it will autonomously go ahead and perform the sanding operation. So that’s the kind of technology that we offer to high-mix manufacturers because before we arrived, there were no such solutions available to them. Mass production people were using robotics, but high-mix manufacturers did not have access to robotics.

Jimmy Carroll: [00:01:49] Yeah. And “high mix” is a term that I’ve seen a lot lately when it comes to the necessity of AI. And we’ll dive into that a little bit later, but I want to keep it at a high level at least at first. S.K., what are some of the major pain points that manufacturers are facing today? And not just labor shortage but including labor shortage?

S.K. Gupta: [00:02:11] So basically there are many different manufacturing operations which are ergonomically challenging, meaning that these are the kind of things that people absolutely do not want to do. Now sanding, grinding, polishing, buffing are all examples of such things. I mean, these things are very hard on your body, so people absolutely don’t want to do. And when you have such tasks that people don’t want to do, then you have a huge labor churn. So that means that you sign up people to do it. You you hire them. But the moment they get trained and they realize how challenging life is going to be, they quit. So basically when you have such large labor churn, it then starts creating all sorts of challenges for you. You’re often working with new people, and new people are more likely to make mistakes. Then you’re going to have a lot of scrap, a lot of rework. Quality is not going to be good. So all these challenges, which are related to quality and extra cost because of rework and scrap, is going to be there as well. In addition you might not be able to meet your deadlines. So these are all challenges that people are facing in manufacturing work nowadays.

Jimmy Carroll: [00:03:30] So yeah, it’s funny, like you mentioned having jobs that people don’t want to do. And I know it’s cliche, right? The dull, dirty, dangerous jobs. But that’s just the nature of it. And it’s funny that,  maybe it was like seven or eight years ago or so, there’s this whole idea of robots are going to take our jobs. And I think we’re sort of past that now. And now it’s robots are doing these jobs because people just don’t want to.

S.K. Gupta: [00:03:59] Absolutely. Yeah. So there are lots of such jobs now where people absolutely would not like to do those jobs, and they’re happy that robots are doing it. Because if robots can do it, then I can add value to the company in a different way.

Jimmy Carroll: [00:04:15] Right. 100%. One thing that stood out to me on your website, and I’ve followed your company for a couple of years now, so I’m actually excited to talk about this and talk to you today. But one thing that stands out to me on your website is this idea that your robots don’t require any programming, and maybe it’s an obvious thing to talk about, but something that a number of companies in the industrial automation space as of late have stressed is ease of use. Obviously, everything should be easier to use, but it’s not always been easier to use. So what does that mean to you? What does ease of use mean? And can you tell us a bit about how you’ve made the fact that your robots don’t require any programming possible?

S.K. Gupta: [00:04:59] So historically, if you want to use a robot, you would need to figure out how to program the robot. And then the robot will repeat the same motion over and over. And that model works for mass production. Even though you may make upfront investments sometime, it may take a couple of weeks to get robot to do the right thing, but hopefully you want to keep doing the same thing over and over for a year, so it’s justifiable that’s why you can do that. Now, if you are dealing with a high-mix application, where the part may change a few times even within a day, or the part variability is so high that if you program the robot, the next part will be different enough and therefore that same program does not work. So obviously, in that case, this whole notion that a human is going to program the robot can take hours or weeks to do it is simply not a useful idea, and it’s not going to be viable. So we need technology which enables a robot to program itself. So that’s the idea. So we want to give a robot the level of autonomy so that if you give it the part, it will look at the part, build the model of the part, and then figure it out: how to move the tool over it. And that’s the capability that we have built in our technology.

Jimmy Carroll: [00:06:28] So I guess this would be a natural time to start asking you about AI. It’s such an omnipresent topic, and it means a lot of different things to a lot of different people. And in the industrial automation space, it’s been subject to overhype. There was a trade show seven or eight years ago in Germany. And these deep learning software packages were being touted as magic bullets that are going to solve everything. And AI is not going to solve everything, but it is an extremely powerful tool that can augment existing industrial automation tools. And it’s kind of settled in in the last however many years as finding its way in industrial automation. So I guess, like for you, what are some ways that AI is benefiting manufacturing today, including your companies but beyond that as well?

S.K. Gupta: [00:07:20] So AI, as you said, is a very powerful tool. And if used appropriately, it can significantly help everyone. So let me first walk through the general uses of AI. And then we will talk a little bit more about how GrayMatter Robotics is using it. So in general, AI is a great tool for automating decision making in a production setting. How do you generate a tool path? What tool path to use? When to schedule a part. So all these operational decisions related to scheduling and planning can be done with AI. The reason why you need to use AI is because these problems tend to be complex and you want to solve them very, very rapidly. You don’t have a significant amount of time to solve them. So therefore you want to use AI tools, which can be helpful in getting you close enough to the optimal solution but get it very, very quickly. So that’s where AI is playing a big role. So figuring out how the tool should move over the part to generate this finish that you’re looking for, how the robot should move the tool. So all of this is good examples of AI being used. Now, in addition to production, you can also use AI in planning your maintenance and service operations, so AI can then give you a prediction that this machine at this time will require maintenance, or you are better off servicing this machine at this time and to overall optimize your production. So AI is also playing a big role in the maintenance and service planning.

S.K. Gupta: [00:09:07] And it can see trends that a human might miss. It can do forecasting of when something is about to go wrong, and therefore it can alert you, and you can take care of things well before it actually occurs. Third big area where AI is playing a big role is supply chain management. So, for example, if you are running an operation and you need to buy a lot of raw material, a lot of tools, and a lot of other consumables, you can basically use AI to manage your entire supply chain. AI can order things for you at the right time. It can get it delivered. It can manage your inventory. It can help you detect if there are any weather events which might disrupt your supply chain. It can issue alerts so that you can go ahead and be ahead of the game and kind of make some decisions and then deal with those challenges. So it’s a great tool in supply chain as well. And finally, if you’re going to be bringing in humans, and you will be bringing in humans because humans are going to do some critical functions for you, so it can also play a huge role in training of humans. So you can use this to train new employees quickly. You can get them certified, you can test their skills. So AI can also play a big role in basically onboarding of all the humans that are going to be employing and integrating in your workforce.

Jimmy Carroll: [00:10:40] So, yeah, that’s interesting. Well, I’m just glad that you brought up a number of different ways that AI can be used. And not just what GrayMatter does specifically, but I do want to ask about that, and “high mix” is a term that sort of set off an alert for me because adding AI into existing or new vision guided robot systems adds so much flexibility. So if you go to a traditional robot, like a so-called blind robot, these robots can perform preprogrammed tasks. You know, it’s the same task over and over. Obviously it’s rigid in that it can only do what it’s programmed to do, By adding 3D vision, that opens up a whole new layer of applications, like bin-picking, for example. It can handle, it can identify, pick, and place parts with fairly high levels of accuracy at this point. But adding AI, that takes it even further, right? So instead of just being able to handle, let’s say, the same plastic or steel automotive part, it can now handle, something in a warehousing environment, for example. If you’re packing for e-commerce or something, all these different sized parts, all at once, without having to go back and program. Is that how GrayMatter is using AI? Are you leveraging AI to be able to have these really high-mix manufacturing environments? It doesn’t require changeover and lots of custom programming and things like that?

S.K. Gupta: [00:12:25] Yeah, so let me just walk you through how GrayMatter uses AI. So the first place where we use AI: you bring the part, put in front of the robot cell. It autonomously figure it out, how to scan the part and build a 3D model. So that’s the first layer where we’ll be using AI. How do you scan it? How do you scan planning, so that you can scan the part in the fastest possible time and then segment it into different regions. So building a model of the part and segmenting it. So then each segment you can process at a time. So that’s the first layer where AI is coming in. And then once you have a segment, that defines the task that needs to be done. So, for example, a patch on a part requires sanding. So in that case AI is also used to generate the tool path, because you have to generate the tool path in real time, and you have to generate the best tool path that you want to use to give you the really good quality.

S.K. Gupta: [00:13:30] So all the path planning of the tool and then the trajectory planning for the robot is also being done by AI. So we are using AI in the beginning to scan the part, build the part model, because we don’t assume that we are given a CAD model because, again, in a lot of high-mix applications, CAD models may not be very accurate because a part might be significantly different from the CAD model. So therefore we don’t assume; we scan the parts ourselves. And then using AI we can actually build the model which has the right fidelity, because we don’t want to be doing scanning super slowly, which builds a very accurate model but takes forever. But on the other hand, we don’t want to scan things so fast that we get a model which is useless and then later on causes a collision. So we want the optimal model, which can be built very quickly and yet be very useful during planning. And the second layer, as I explained it to you, is all about then planning the tool motion and planning the robot motion.

S.K. Gupta: [00:14:34] Now in the case of sanding, you can also monitor the tool wear. So you really want to see how the sandpaper is wearing out, and you want to then figure out when to change it. So AI is also monitoring the sandpaper and basically making sure that we change the sandpaper at the appropriate time. We extract the maximum life of the sandpaper yet we manage to change it in time so that the quality remains good. And then AI is also being used to basically monitor your operation. So what if somebody forgets to clamp the cord, and therefore the moment you begin the sanding process, the parts start moving, right? So AI is also needed to make sure that your operation is going as planned. So you’re basically getting pretty much expected responses from sensors, and if physical sensors are not giving you the response which your virtual sensors are telling you what you should be getting, then you can detect anomalies and then classify what kind of problem it might be. Is it that somebody put in the wrong material? Did somebody’s load the wrong type of sandpaper? Is the part not properly clamped? There’s some other issues where TCP, if the tool is not calibrated right. So we can also predict what kind of problems are there. So we call this technology basically GMR-Guardian. So that ensures that your process is progressing smoothly. And if there are any anomalies being detected, then we can bring the system to a safe state and pause and issue an alert. So that’s another way by which we use AI technology.

S.K. Gupta: [00:16:32] So we use AI quite a bit in different settings, right from the very beginning of the scanning to all of our operation planning and then managing all of our sandpaper changes or overall in general basically consumable management, and then finally monitoring the process when it’s being executed and ensuring that we can then detect any issues and then basically intervene and make sure the process delivers the quality.

Jimmy Carroll: [00:17:05] Yeah, very holistic approach to AI. I love it. I do want to ask a couple more. I don’t want to linger too long on AI, and I’m sure you don’t want me to either. But I do have a few questions that I’d like to ask. One of them is, on the inspection or vision guidance part of it, what is GrayMatter’s approach to training datasets? Like where do you bring, as far as human in the loop, labeling images. Do you ask your clients to do that for you? Do you work with them on it? Do you do you use synthetic data generation? What’s your approach there?

S.K. Gupta: [00:17:39] So we don’t ask our clients to label the data. However, we ask our clients to ship us some parts, and then we run the experiments in our facility, and we do some manual labeling of the data. So that is done by our engineers. So we do that, and we also use synthetic data. So a combination of labeling the images in our facility and as well as synthetic data. So that’s kind of how we generate the training dataset.

Jimmy Carroll: [00:18:12] Okay interesting. Let’s see, I wanted to ask about sustainability. So industrial automation, and within that we’ll say AI is part of industrial automation technologies. I’ve seen a lot of larger companies lately promoting the fact that these technologies promote sustainability, and maybe some of the ways are obvious. But to you, what are some of the ways that these technologies help promote sustainability for manufacturers around the globe?

S.K. Gupta: [00:18:42] Yeah, this is a great question. So let me dig deeper into it. And I can tell you something which is happening in the near term and something that we expect should happen in slightly longer term. So near term, what ends up happening is that you can have a huge saving on consumables. So, for example, a robot can extract the full use out of sandpaper, while a human, when the sandpaper starts wearing, may feel that they have to apply now extra force and therefore they do not want to do that because that’s going to hurt their arm, and therefore they’re not going to be interested in getting the last bit of usefulness out of the sandpaper. So they might prematurely throw it out. So we have shown that a robot can actually significantly reduce the sandpaper use. So that’s a good example of just reduction in waste and consumables. So it’s one way AI-powered robots are helping. The second aspect is that you can dramatically reduce rework and scrap. So rework and scrap means that all energy that was spent in doing the operation goes to waste, because you have to rework it or you’re going to scrap it. So robots again can give you very consistent performance, and they can dramatically reduce scrap and rework, and that has, again, savings from a consumable perspective also savings from an energy perspective.

S.K. Gupta: [00:20:23] And also saving things from going to the landfill. So that’s all the positive benefits of just having significantly better quality. Another interesting dimension is a lot of processes that I mentioned, people have to wear PPE. So the people are not there, then all the PPE, which is going to the landfill, also doesn’t basically go to the landfill. So obviously we save on that as well. Couple of other interesting things which are beginning to emerge. One is that if you can get robots to do sanding, then you could probably maintain a building temperature between 60 and 90 degrees, right? So in winter you can set the thermostat at 60 if robot is doing things, and in summer you can set it at 90 if robot is doing things. Now, if humans are doing things, you have to make the workplace a lot more comfortable. Now just the saving that can come from building energy consumption point of view can be huge. So that’s another dimension through which AI-powered robots are beginning to play a role. Now as one imagines a future where more and more AI-powered robots will be used, then people could do more remote work. They can manage the robot. They can do some quality assurance type of work from home.

S.K. Gupta: [00:21:57] So people are also forecasting that this will reduce the need for people to commute. And that also has interesting implications from an energy use perspective. So overall AI-powered robots can have some very direct impact in terms of consumable reduction, PPE reduction, scrap production. All of that has the benefit of reducing things which are going to the landfill, and if you have significantly consistent quality, then you can also lower your energy consumption. And people are also coming up with all kinds of innovative process recipes. Once robots go to work, you can apply more force, you can go faster, you can do many more interesting things. So overall, once you are able to go beyond human capabilities, then you can also make the process inherently energy efficient. So there is a significant gain on the energy aspect as well. So that’s the kind of summary of what the direct benefits are. And then the indirect benefits are that perhaps you can save significant energy in HVAC, reduction in the energy, that particular part of it, and also the need for people to commute to work. If more people can do remote work, then that has energy saving implications as well. So that’s kind of a spectrum of possibilities that people are beginning to explore.

Jimmy Carroll: [00:23:31] Yeah, sure. No, I love that. And it kind of ties in nicely to another question I wanted to ask. It stood out to me a bit, and I love it. Your mission is to enhance human productivity and improve quality of life. So like I mentioned it earlier, on one hand, it seems like we’re getting over the fear, people’s fear of robots stealing their jobs. But on the other, now there are some people out there who are who are scared of AI. And I’ve seen the movies, and for people that maybe are outside in the general public, it might seem scarier than it actually is. So, one, how do you feel your technology and automation technology in general can improve quality of life? And then, two, how do you dispel the notion that AI is scary?

S.K. Gupta: [00:24:19] So basically, first let’s talk about quality of life issues. So AI-powered robots or autonomous robots are helping the workers because, again, workers don’t have to now take on these ergonomically challenging tasks, which may pose health risks for them. So obviously workers who are doing sanding, grinding, polishing, their life will be better if they were using AI-powered robots. Now people then often say, okay, what role workers could play? Now remember, once you have a robot, you need to develop new sanding recipes. You know, hopefully robots can apply much larger force, can move much faster, and therefore one can invent a new recipe. So people who have years of experience in process could take on those roles. Can I invent a new recipe? Can I get a robot to do better? Because again, the robot need not necessarily copy the human, right? The robot can do things in a different way. So all kinds of interesting possibilities are there. So that’s the impact on worker perspective. Now, if you look at the consumer perspective, from the consumer perspective we are getting basically much better quality products. We can basically get things which are much more consistent.

S.K. Gupta: [00:25:53] And obviously the cost goes down. Then everybody benefits in terms of products being more affordable. And if you look at the society more broadly, because of all the sustainability thing that I mentioned, there is a tremendous benefit to society as well, not just the workers and consumers but also overall society benefits, because if manufacturing is more sustainable, then all of us benefit from that. So that’s kind of a quality of life and productivity piece, right? Where this is going to be good for people. Now let’s tackle the AI piece. So from my perspective, modern AI is able to query a vast amount of stored data. And from that data can make some good recommendations. So if you boil it down to the training part of AI and then use part of the AI, that’s what AI is all about, right? So you have a stored massive amount of data, and then you have trained it, which basically allows you to put the data in a different view, and then you just query it.

S.K. Gupta: [00:27:13] So I don’t think people’s notion of AI taking over the world, reality is far, far away from it. I mean, this is just basically a massive amount of data stored, reformatted, and then during runtime just query it. That’s what most of the AI is all about right now. So it definitely enables computers to make complex decisions faster. But it’s not going to be taking over the world or basically creating all sorts of other doomsday scenarios that people are currently thinking about.

Jimmy Carroll: [00:28:01] Yeah, yeah, that’s the thing, right? There’s some of the mainstream uses of AI, like generative AI for images, which can get nefarious, or large language models, ChatGPT and things like this. People see these and then they see the movies and the TV shows like Black Mirror and all this. I have friends that are very intelligent that truly believe AI is going to take over and we’re all doomed. And I’m like, no, you don’t understand. Even generative AI right now. ChatGPT. I had a podcast a couple of weeks ago with a guy from a systems integration company, and he was a director of AI, and he said, you know, on one hand, generative AI is very useful. It’s finding uses in the industrial space for things like making PLC programming easier. But on the other hand, with all the hallucinations that take place, do you fully trust it? And he cited this example that he gave me, and he said, I asked I asked ChatGPT recently to give me the names of the Snow White’s 13 dwarves, and instead of realizing that that’s not right, there’s only seven – I think seven, if my Disney history is correct – but it made up names that sounded like the other names of the dwarves, because it just is not there yet. So I think that people have to maybe pump the brakes a bit when it comes to mainstream AI. But what’s your take on generative AI? Do you think it has any use in industrial applications?

S.K. Gupta: [00:29:41] Absolutely. Lots of use. So, you know, basically a lot of synthetic data that need to be generated. Generative AI is going to play a huge role. Sometime when you need to reason about the world and your sensing has left some holes in it, generative AI can fill those holes. So that’s another great example. Code generation, we already talked about it.

S.K. Gupta: [00:30:08] Simple software code now can be generated by generative AI, which is another good example where you want to integrate output of this particular system with that particular system. Why should a human waste a lot of time writing code when you can get generative AI to write that routine kind of code? Generative AI is also great for generating test scenarios, which then can be used to test your system, and that way you get a much more robust system. So yeah, so generative AI, huge role in synthetic data generation, huge role in filling the sensor data gap. Huge role in writing some routine programs and also generating test scenarios and test cases which can be used to rigorously test the system. So all of these are great examples. And we are using generative AI for all of this.

Jimmy Carroll: [00:31:03] Yeah, absolutely. That’s great. A lot of examples of AI to talk about. And I could keep going on it, but but I won’t. You’ve recently received some funding, and in general you’ve got an impressive list of investors on the website. So what’s most exciting for GrayMatter right now? What’s next for you all?

S.K. Gupta: [00:31:26] So we keep expanding our line of products that we offer, and we keep making our technology more and more robust and get it to perform better and better. So that requires some R&D. That requires a lot of capability to serve our customers better. So we are growing our team and offering new products and improving existing products. So all kind of exciting things that are happening in that space.

Jimmy Carroll: [00:31:58] Absolutely. What else are you excited about? I guess in the larger industrial automation space, anything come to mind?

S.K. Gupta: [00:32:06] I think overall the thing that I’m most excited about is that everybody is starting by saying, okay, here is a process and I cannot find humans to do it. Let a robot do it. But oftentimes what happens is that the moment human constraints are eliminated, you can truly innovate. You can truly do things which are different, which are creative. Because once you’re no longer confined to the human form factor and all the limitations that come with the humans. Humans are great, but all of us have a lot of limitation in terms of power that we can supply, the force that we can apply, motion or the speed at which we can move things, and our accuracy. So our power, our force application ability, our speeds, and our accuracies are very limiting. So the moment these human constraints get eliminated, the world opens up, and then suddenly you can innovate and you can actually come up with a dramatically different, better way of doing things. And that’s where the biggest payoff is going to be. And then we can innovate. We can really do things in a significantly better way, as opposed to saying, okay, here’s how the human is doing it. Let the robot replicate it. So if we get past that, and we can simply say, okay, humans are doing it now, let’s figure it out. What would be the optimal way for an automated process to work? That would be a great thing, because then we can speed up in many sectors manufacturing by a factor of five, factor of 10. That adds a lot of capacity, can also have huge ramifications for sustainability. So I think all kind of exciting things can happen once we start viewing automation  basically as a tool for innovation. Meaning that this can actually now do a lot of innovation, because typically people have used automation as a cost-cutting measure or cost-containment measure or dealing with labor shortage measure. But automation has huge potential for innovation. And that then creates new types of products that can make the work a lot more sustainable. So I think viewing automation as enabler of innovation is what I’m most excited about.

Jimmy Carroll: [00:34:49] Yeah, I love that. And I promised I wouldn’t bring up AI again, but I’m going to because it reminds me of something that Andrew Ng said. Andrew Ng is a founder of Coursera and Landing AI, and he was a member of Google Brain and a prominent member of the AI community obviously. And he said that AI is the new electricity. And if you think about that and you think about all the new doors that the combination of AI and existing industrial automation technologies like robots and machine vision and everything that goes into them will create, it’s very exciting. And it’s interesting because we could be on the brink of a new manufacturing renaissance type of thing. Anyway, it’s interesting to think about. Do you have any thoughts on that?

S.K. Gupta: [00:35:36] Yeah, definitely. So AI is going to be a pervasive tool. Everybody will use it. It’s just a tool. So that’s the right way to think about it. It is a tool available to all of us. And if we use it the right way, it’s going to have a huge win for everyone, right? So everybody wins in that case. And it delivers much better products, makes things a lot more affordable, makes things a lot more sustainable, and it creates a lot many new business opportunities. So it would be great if we use it in a responsible, right way.

Jimmy Carroll: [00:36:11] Absolutely. Yeah. And then sometimes in some cases it can be fun. Like some of the mainstream examples we talked about. It can be fun. It doesn’t have to be scary for everybody. Well, S.K., thank you so much for taking the time today. I really appreciate it. If anybody has questions or would like to reach out to S.K., they can go to Graymatter-robotics.com, find them on LinkedIn, find S.K. on LinkedIn, or if you’d like to pass questions to us, we’d be happy to pass them along. It’s manufacturing-matters.com. So S.K., once again, thank you very much. It’s been a pleasure.

S.K. Gupta: [00:36:45] Thank you for inviting me. I really enjoyed this conversation. And yeah, I’m very, very happy to answer any follow-up questions that people may have. So once again, thank you very much.

Jimmy Carroll: [00:36:57] Of course.