Smarter Warehouses - AI and Robotics at the Forefront of Logistics Innovation

- Welcome to another episode of "Supply Chain Frontiers," the MIT CTL podcast where we explore the latest trends and innovations in supply chain management. I'm your host, Benjy Kantor, and today, I'm thrilled to welcome back Dr. Matthias Winkenbach MIT CTLs Director of Research, and more pointedly, with regard to this episode, the Director of the brand new MIT Intelligent Logistics Systems Lab. Matthias has been at the forefront of advancing logistics through innovative technology, and today, we'll delve deeper into the partnership with Mecalux, a provider of warehouse management solutions and a founding research partner of the Intelligent Logistics Systems Lab. First, this episode of MIT "Supply Chain Frontiers" is brought to you by the MIT Center for Transportation and Logistics Executive Education Program. Offered twice a year, January and June, this program's designed for corporate leaders and aspiring executives in supply chain, logistics, procurement, AI, and related fields. It's open to learners at every level focusing on strategic skills and industry insights that drive innovation and growth. Instructors include experts from CTL and across MIT. Our next executive ed session is right around the corner. To learn more, visit ctl.mit.edu/executive-education. Joining us on "Supply Chain Frontiers" today are Alejandro González, the Software Business Unit Director at Mecalux, and his colleague, Mecalux Chief Technology Officer, Iñaki Fernández, who are here to share insights on how this collaboration aims to transform logistics systems and improve efficiency. We'll discuss the future of intelligent logistics and how companies like Mecalux are leveraging AI to enhance supply chain operations. Let's jump right in. Well, thank you for joining me, everybody. I really appreciate it. Just to kind of follow up on what we discussed a couple months back, Matthias, could you give a recap of the purpose of the ILS lab and how it came to be?

- Sure, yeah, so I think as we discussed last time when we spoke on this podcast, we've been doing research in the logistics and supply chain industry for decades at CTL, and we've mostly done it from an operations research point of view. So we used optimization and simulation methods for all sorts of different projects we're embarking on with our research partners. And lately, there's a lot of hype, obviously, around machine learning and artificial intelligence, and sometimes you get the feeling that people are just trying to put a little bit of the AI sprinkle on everything to make it more interesting. And that's exactly what we did not want to do. We wanted to actually have a lab that works with a credible industry partner or actually a set of credible industry partners on identifying actual use cases, actual fields where AI can augment what we've been doing so far with purely operations research focused methods. So trying to identify use cases in the logistics and supply chain industry where AI and machine learning are not just a buzzword but actually create additional value. And that's a great fortune for us, that with Mecalux, we found a founding partner who gave us the means to actually explore this, because some of this boils down to doing very fundamental research at first, research that may not immediately yield commercially viable solutions, but research that's necessary to then, in subsequent iterations of this collaboration, yield to solutions that can actually be applied.

- And before we bring in Alejandro and Iñaki, what prompted the collaboration specifically with Mecalux?

- Well, I mean, what prompted it was that actually the CEO of Mecalux was listening to a podcast of Mecalux that I was a guest on a couple of months ago by now, and I kind of expressed that idea that rather than just following the hype, we would want to see what are the most useful application areas of AI and machine learning in our industry. And I think that kind of got him excited. That's why Mecalux reached out to us to see whether we would be interested in a research partnership. And obviously, we were. Fast forward to today, we are here. We just kicked off this collaboration formally and we're working on the first research project.

- The power of podcasts.

- Yes.

- Can you believe it. Well, Alejandro and Iñaki, from the perspective of Mecalux, what do you see as the key benefits of the partnership here? Why would a company like yours get involved with CTL and the work that Matthias and his lab are doing for your company and your clients and the world at large?

- At Mecalux, we've been in the logistics business for a long time, so we have a very deep understanding of how things work. We believe that by combining MIT's academic knowledge with our own practical real world experience, we aim to develop groundbreaking research that will make a real difference for our clients.

- Yes, and to compliment to what Iñaki was saying, I think that one of the goals is also to take these breakthroughs in AI and robotics for example, and roll out them across our customers warehouses around the world, and obviously, in a quick and efficient way. So we're really focused on trying to deliver tangible results that our customers can benefit from.

- Great, I understand at the launch of this, and Matthias and I were talking about this in our last session, but the lab has three simultaneous research projects. I know it's very early, so to ask what's been done already might be a little premature, but what are the next steps there, Matthias?

- Yeah, I mean obviously we are in this exploratory phase right now. We just started working on this a couple of weeks ago. But broadly speaking, we are looking at two fields that have an immediate impact, not just on Mecalux's business but also have an immediate commercial impact on any large scale logistics and supply chain company out there. One, is looking at warehouse robotics. Warehouse robotics is an area where AI and machine learning can actually, more or less, immediately be applied to the real world. Unlike areas like autonomous cars or drones or what have you, this is not an area where we're talking about applications that may or may not materialize in 5 or 10 years. This is an area of application where if we find a better solution, we can actually try it out today in a real operation. So what we are looking at there is trying to make autonomous robots that move around in warehouses to, for instance, transport boxes or pallets from A to B within the four walls of a warehouse, trying to make those robots smarter. In that sense, also make them work more efficiently together with human operators. So make them more robust, for example, against random interruptions. Think of a human warehouse worker walking in front of an autonomous robot. Right now, that might shut down the robot and basically harm the productivity of that entire system. We're trying to build algorithms with which these robots can navigate in a more human-like, more intelligent way to basically increase overall productivity of that warehouse system. The other area of application is more related to software solutions that help companies manage demand out of a distributed network. So what does that mean? So if you're for instance, a retailer and you have a large, potentially global network of warehouses, fulfillment centers, retail stores, all of which can theoretically fulfill orders, for instance, of your online customers. So as soon as someone places an order online, just to stay within that example, you have to make a choice. Like where do you fulfill that order from and when do you fulfill it? In which priority order do you fulfill orders? And related to that is how do you manage inventory? Where do you keep your inventory in such a distributed network such that you can actually serve your customers most reliably and most cost efficiently? So these are very complex optimization problems to which our traditional domains or operations research has only found, let's say, approximate solutions. We're hoping to combine those existing methodological approaches with newer methods, machine learning methods that actually make them smarter, make them more data-driven, help us make these decisions in a better, more efficient way. And the benefit of that could be manyfold. It could obviously be commercial benefits like reduced costs, but also social and environmental benefits. So for instance, the better we manage this, the less unnecessary transportation might be happening in our network. So the fewer unnecessary emissions. And similarly, if we manage this really well, we can reduce waste. So there's many possible benefits of doing this right. And that's why we are so excited about working on this.

- The first part of your answer was reminding me of a story from a presentation from one of our colleagues, Dr. Miguel Rodriguez Garcia, who works with our MicroMasters program, who also works with robots and warehouses. And he was telling a story about how this completely efficiently built warehouse where two robots smashed into each other, started a fire and the entire warehouse burned down. So other than that sort of practical, wanting to avoid that kind of situation, what do you see, Iñaki and Alejandro, as the key benefits for this partnership? Why do these projects matter to Mecalux?

- So they three research projects that we will develop together with MIT are key for us. So when it comes to AMRs, AMRs stands for autonomous mobile robots. They are becoming increasingly popular in warehouses, to the flexibility they provide. They make it easy to scale operations and respond to changes and demand by adding more robots. So this design of research with MIT will help us create a new generation of AI-powered AMRs that outpace the performance of today's market options. With reinforcement learning, we can actually maximize this machine's productivity in warehouses and actually driving greater efficiency.

- Then going to the software side of things, at Mecalux, part of our DNA, it's obviously innovation. So we really believe that innovation is extremely needed to be able to provide the best solutions for our customers. And this is why we have a large team of engineers at this moment working constantly on innovating and improving the capabilities of our logistics software. And through the joint software focus research lines, in the end, we aim to harness AI's potential so that the models can self-train and learn from past experiences. And this self-improving AI will ensure, for sure, a maximum productivity in warehouses and will integrate these breakthroughs into our software. So helping, in the end, our companies to establish distribution strategies that dynamically adapt to the environmental demands through the AI.

- Well, you've mentioned technology and innovation a few times, and with that kind of in mind, Matthias, I guess this question is for you, how is the lab initially leveraging AI and machine learning in the research process?

- So we obviously can't go into the methodological details here, but the general idea is that we're trying to improve our systems that already exist. Both those autonomous mobile robots that we've been talking about, as well as the software that manages, for instance, online orders in a network. These solutions already exist. Today, they are performing predominantly based on predefined rules. So some human developer at some point had a good idea and wrote down a bunch of rules by which, for instance, the robots decide how to navigate through a warehouse or by which distributed order management software decides which facility should fulfill an order. And these rules are generally doing a decent job, but they're not perfect. So the first step that we're trying to achieve here is we're trying to calibrate our solution against the status quo. So in a way, we're trying to imitate the behavior of those human-defined rules in a learning-based environment. The first step is just trying to establish a machine learning model, for instance, that can more or less replicate what the current systems are doing, because that's kind of our performance baseline, what we want to compare ourselves against. Then in the second step, we try to improve over this. So Iñaki mentioned reinforcement learning. There's a variety of different methodological approaches that we could choose here, reinforcement learning is one of them, but the idea basically is start off that baseline that mimics the status quo and try to make these systems improve over time. So for instance, we try to make the robot learn from experience and learn from previous navigational choices that might have led to suboptimal outcomes. So for instance, it chose to turn left, but unfortunately got stuck in a cul-de-sac, for example. But giving those machines the ability actually to reflect on their behavior, reflect on the performance that they get from it, and try to improve next time.

- Like a parent, I'm talking to my 6-year-old, and I said, "I want you to think about what you've done."

- And kind of the other nice concept behind this is that if we are in a rule-based world, whoever designed those rules, designed them with a certain objective in mind, but was actually never really able to kind of prove that the rules that that person defined would actually, for instance, lead to the optimal cost performance of the system. Now that we're in a learning-based environment, we can actually define what our objective is and that objective might also be much more complex than just minimizing cost or maximizing lead time or whatever. We can basically define whatever objective we wanna have and let those systems learn the best behavior, the best policy to achieve the best possible outcome along that objective. And that's something that existing methods basically can't really do and that's why we're using machine learning methods to add that capability.

- And are there existing technologies that Mecalux is already using that you're integrating into your research or planning to?

- Well, in a sense. So as I said before, we are trying to mimic the status quo first. So for instance, to stick with the robots, we're obviously using the robot solutions that Mecalux is building and selling to its customers as kind of the technology baseline. So these are the vehicles that we're trying to work with. These are the vehicles that we're trying to model and simulate and hopefully, improve. And we're also obviously using the existing software that makes these machines think as our baseline. So the current rules by which these machines navigate through a warehouse, that's what we are building off of. In the long run, obviously, some of our research may in turn be reintegrated into these machines if we can show that our methods, our models can actually outperform what is currently being done.

- Similarly for you, Iñaki and Alejandro, is Mecalux thinking so far ahead in the idea of the findings of this research actually being integrated into Mecalux business solutions and what is the thinking beyond that?

- Yeah, so short answer is yes, but let me develop that a little bit. So while our software automation and robotic systems are constantly evolving because obviously, we need to follow what our customers in the end need, we aim to take them to a step farther by integrating MIT's discoveries into our technology. And at Mecalux, we are aware to stay at the forefront of logistics innovation, it is obviously essential to continuously adopt the breakthroughs in the science and engineering. And this is why we believe that MIT's pioneering research will help us to develop a new generation of logistics technologies that go beyond just efficiency and productivity. So they will truly transform how goods are stored, moved, and managed, and not only within the warehouse but also outside the warehouse and more talking about supply chain networks, which is also one of the topics that we wanna address.

- So we have high expectations on this collaboration. We hope to fully leverage AI and machine learning to give our customers a significant competitive advantage. With MIT's support, we will be able to push the limit of these technologies from finding demand forecasting to developing smarter, faster warehousing operations. So integrating research outcomes into our product will, I'm sure, it allow us to offer advanced predictabilities that anticipate our customers' challenges and needs.

- And I'm glad you brought up that point 'cause it's an interesting challenge from a practical standpoint. What are the biggest challenges? How do you actually practically integrate this research and advanced technologies in general into your traditional logistics processes?

- So one of the biggest challenges that we face in integrating advanced technologies into our traditional logistics is the balance that always exists between innovation and practicality. So many warehouses still rely on long established processes, sometimes even manual ones, and that have served them for decades. But bringing new technology like automation, like advanced warehouse management systems, dumb softwares with sophistication of order orchestration, inventory optimization, all of this together obviously requires a mindset shift, and sometimes a complete rethinking of the workflows of the warehouse and how their supply chain operates. So this is one challenge then, another one is adaptation. It's also one of the challenges that we most commonly see in our customers, because for some clients, the shift from traditional methods to something like what we are trying to do in this program, which is AI-driven inventory management can seem a little bit daunting, but at Mecalux-

- Just a little, just a little bit. Yeah.

- Yeah. But at Mecalux, we try to invest heavily in designing user-friendly interfaces and providing hands-on training and support to our customers to help them to ease this transition throughout the years and throughout the use of our technology. And last but not least, I think that as technology keeps progressing, so do the expectations around speed, accuracy, and data insights. So we are constantly innovating and refining our solutions to stay ahead of market demands. For example, we are focused on making our systems more adaptable, scalable but responsive at the same time so that they can grow with our clients' needs and future proof of their operations.

- What are the hesitations for adopting new technologies within supply chains? Is that something Mecalux experienced, like a hesitation or a sensitivity to it? If so, how do you overcome a resistance, I guess, internally with staff from the top all the way down with communicating that kind of decision to overcome the resistance to integrating new technologies?

- That's a great question, Benjy. There are several reasons why organizations might be hesitant to adapt new technologies within their supply chains. But one of the most common reasons is simply challenge of change. Especially in the field like logistics where traditional methods have been reliable for decades, we have many customers worried that adopting new technology will disrupt their operations or lead to costly downtime. At Mecalux, we have systems in place to ensure a smooth transition to new technology. For example, one of our recent projects, we automated IKEA's components warehouse without disrupting ongoing operations. Their specific requirement was to implement automated systems without altering the warehouse structure or holding operations. That was a mass requirement. And the project was a success and has boosted logistics efficiency. Now, they fulfill 99% of the orders on time and in full thanks to automation.

- To build on that, I think what it essentially boils down to is trust. And that's probably the biggest challenge when you think about machine learning and artificial intelligence compared to traditional methods that have been used in the supply chain and logistics industry. Because at the end of the day, most machine learning methods are a black box, not just to practitioners, even to researchers. We don't exactly know how such a model will respond. So to give a very tangible example, if you have a robot moving around the warehouse, and so far it's been controlled by a set of rules. One of those rules might be if you get closer than half a meter from something else, slow down. And you can see that rule in the code of the software that controls that robot, so you know exactly that that's the behavior that you're going to see if that robot comes close to something that might be dangerous. In the machine learning model, doing the exact same thing, you don't see that anymore. There is no if then else statement that says, "If X happens then do Y." You're basically trusting that the model learned that this is the desired behavior, but you have no guarantee, black and white, in code, that would tell you this is exactly how the robot is going to behave. So this is just a made up simple example, but it comes down to whether it's safety, whether it's business continuity, whether it's all sorts of risk associated to integrating so-called smart systems in a logistics network, in a logistics facility. You have to build trust that it will behave the way that it's intended to behave. And that is probably the biggest hurdle that we currently see from a research point of view, not just with Mecalux but with all of our research, that this transition towards models and methods that are no longer human explainable, necessarily, is a big mental shift.

- With your example, this is maybe perhaps getting in the weeds a little bit, but how do you build in that quality assurance where, like if I'm just using generative AI to help me write a paper or write an article, I can then read through it pretty quickly, see where the corrections need to be made and that, but if you're talking about the safety issue where you're expecting that the robot knows that it's not supposed to go, how do you test that? How do you, in a real life scenario?

- Yeah, I mean ideally, you wanna not test it in a real life scenario first.

- Yeah, good point.

- Even though that's where we wanna go to, obviously. But, so for instance, one of the first things that we are currently doing in this research collaboration is building out a very detailed simulation test bed so that we can actually mimic the real world operations of any warehouse that we might have out there virtually, build almost like a virtual twin of a warehouse so that we can throw any kind of new algorithm at that and see how, in this case, the robots would behave. Obviously, the challenge here is to model the simulation test bed as accurately as possible so that we can depict all the possible things that might go wrong in the real world. And to be honest, we'll never get there, we'll never get to 100% of the true complexity of a warehouse. But we wanna get to, let's say a confidence level where we're like, okay, we are sure enough about this algorithm, now, we can actually test it in the real world. Now, we can actually deploy this to a pilot warehouse of some customer and see what happens. And obviously, we wanna start small there, test it out at a small scale so that if something doesn't behave the way it should, we can immediately intervene and the damage is limited. But that's typically how this goes. You first virtually test it, then you do a small scale real world pilot, and then hopefully, everything pans out and you can roll it out at scale.

- And I don't think I'm late to the game in saying this theme of the importance of trust when it comes to AI is gonna be just a growing and growing theme. And probably, it is already in academia, but like I know that our colleague, Bryan Reamer, with the Advanced Vehicle Technology Consortium has been talking about, like this is the only way that we're gonna get movement on this, when there is public trust. But it seems like there is some level of corporate trust here on Mecalux's side for when you're putting forth that, and high expectations for when the Intelligent Logistics Assistance Lab is producing some research. So I guess for all of you, given that the trust exists there or that you're trying to establish this, how do you envision this partnership between CTL and the lab and Mecalux influencing the future landscape of supply chain management, of warehouse management, things like that?

- So ultimately, our mission is to help, as I said before, companies to achieve operational excellence. Whether it's AI, robotics, or smart warehouse management systems, or even distributed order management systems, we're constantly looking for new ways to help our clients to optimize their operations and therefore be able to better utilize their resources and save time and cost, which is in the end, the main goal of every company. So by this collaboration with MIT, what we wanna do is that we want to push the boundaries of what's possible in the logistics technology. And we envision that this partnership leading to breakthroughs in areas like AI-driven automation, predictive analytics, and smart inter-logistics robots, which we can then integrate into our solutions to help in the end, companies to operate with a whole new level of efficiency and productivity.

- We strongly believe that this research partnership with MIT and with Matthias' team will bring in the best and most advanced technologies to help managers make logistics processes more efficient. For example, with AI-enabled systems, we can empower clients to anticipate issues before they arise. They will be able to optimize resource allocation in real time and and make faster data-driven decisions.

- If I may add to this.

- Please.

- So obviously, you might ask, well, we have, to some extent at least, self-driving cars out there on the road. We have companies that fly rockets to space using machine learning and AI to some extent, I would assume. So why does the logistics industry still not know how to operate robots intelligently? And I think the answer to that is that the logistics and supply chain industry is notoriously risk averse, which kind of relates to what we discussed earlier about that trust issue. And also supply chain logistics traditionally has been very much focused on cost, on very short term returns on investments. And that's honestly one of the main challenges that we were facing in recent years when we were looking for partners to actually help us get this type of research off the ground. Because it's not like the industry didn't see the potential value of these methods, but the fact that it might take a couple of years until we actually see solutions emerge from this that really turn the needle in terms of cost efficiency or whatever the metric might be. And the fact that we're talking about methods that may, at least initially, behave differently from what we were expecting them to behave like, that usually turned a lot of traditional logistics, supply chain management companies away or basically made them not interested in investing into research and development in this space. And that's why for us, it's such a fortune that with Mecalux we found a partner who was willing to take that leap of faith, both in terms of trusting the solution and in terms of allowing us to experiment, allowing us to build the foundations without an immediate need for a positive return on that investment.

- Well, let's talk a little bit about that investment or those expectations, I guess, which is, my understanding is that this is an initial five-year partnership between CTL and the lab and Mecalux. What are the hopes for the benchmarks that you will see over those five years? I don't know how concrete or ethereal they are. Is it after year one, we're going to have X, after year two, we're gonna have Y, that kind of thing?

- I mean that's hard to predict. Honestly, my main hope for this is, first of all, that this research collaboration will be a little bit of a trailblazer, if that's the right English word to use, I don't know, but-

- Makes sense to me. I've spoken English all my life.

- By the means given to us through this partnership, we can show that AI and machine learning is not just a buzzword, a little bit of sprinkle, and at the end of the day, it's overly hyped, but that we can actually come up with solutions that are trustworthy and make economic sense, make practical sense, make also sense from an environmental, social point of view. So basically make sense along all the possible dimensions along which we want to optimize the future logistics systems out there. And once we can show this, we hope that this resistance that I was describing earlier of the logistics and supply chain industry as a whole towards this type of innovation will slowly reduce and we'll see more and more engagement from a larger part of the industry in academic research that might hopefully bring the logistics industry back on track compared to other industries. And the other thing that very practically that I hope to see from this partnership is that we get to try out things in the real world. Because very often in academia, you are a little bit in a weird situation where you come up with super cool methods, super powerful models, you think you invented the best thing since sliced bread, and then no one is willing to actually try it out in real life. And here this partnership, I think we are at a point where we want to try this out in the real world. And that's, for me, I would be very happy if let's say, in a year from now, we would see some robots moving around the warehouse that are actually powered by some of the work that we do here.

- I'm sure we will.

- Yeah, I guess for Mecalux's guess, like how do you see the role of what's happening, not just here, but here in terms of the research on the logistics side of things over the next five years, but the other things that Mecalux is investing in. Like how do you see the advent of new technologies being part of what you're doing over the next few years?

- So over the next 5 to 10 years, we see logistics continuing to transform from a traditional support function into a strategic driver of business growth. Technologies like AI, machine learning, productive analytics will play a huge role on this. They will enable more precise demand forecasting, optimize inventory management, and actually smarter decision-making across the supply chain.

- And adding something to what Iñaki has already said, so we also foresee strong trends in the market towards greater automation and robots in warehouses. So automation will no longer be just an option for large scale operations, especially when we're talking about smart inter-logistics robots like the AMRs. So this is a really good way for companies to make their first step with automation, probably with affordable investments. So it's gonna become more accessible and essential for businesses of all sizes. And what we've seen with labor shortages and the need of faster turnaround times, automation and digitization and also the application of AI in all of this will for sure play a key role and it's gonna be really, really important to meet customer expectations in terms of speed, accuracy, and flexibility.

- Thank you, this has been the 31st episode of "Supply Chain Frontiers." We appreciate you joining us. We appreciate the insights shared today by Dr. Matthias Winkenbach of MIT CTL, and our guests, Alejandro González and Iñaki Fernández from Mecalux. As we continue to explore the intersection of technology in logistics, it's clear that partnerships like these are paving the way for a smarter and more efficient supply chain landscape. I'm not only fortunate enough to host "Supply Chain Frontiers," but on this episode, I'm the producer as well. Our sound editor is Dave Lashinsky of David Benjamin Sound. Our audio engineer today is Kurt Schneider of MIT Audio Visual Services. MIT "Supply Chain Frontiers" is recorded on the MIT campus in Cambridge, Massachusetts. Be sure to check out our previous episode with Matthias for more context on this partnership. And stay tuned for our next episode where we'll continue to delve into the innovations shaping the future of supply chain management. You can hear all episodes at ctl.mit.edu/podcasts or search "Supply Chain Frontiers" on your preferred pod host. Until next time, on behalf of the researchers, instructors, staff and partners here at the MIT Center for Transportation and Logistics, I'm Benjy Kantor and thank you for listening.