Product-driven solutions highlight the practical side of machine learning and AI
For the latest episode of our manufacturing podcast, Data In Depth, we sat down with Bastiane Huang with OSARO. Bastiane digs into the practical side of machine learning and deep learning and details some product-driven, real-world solutions using AI. She details how machine learning can be used to AI-enable robots for less structured environments.
Check out this episode below and head over to Data in Depth to listen to our other episodes!
Have a topic you'd love to hear more about? Interested in being our next guest? Let us know in the comments below!
Announcer: Hi, and welcome to data in Depth podcast where we delve into advanced analytics, business intelligence, and machine learning, and how they're revolutionizing the manufacturing sector. Each episode, we share new ideas and best practices, to help you put your business data to work. From the shop floor to the back office. From optimizing supply chains to customer experience. The factory of the future runs on data.
Andrew Rieser: Welcome, and thanks for joining us, for season two of Data in Depth, the podcast exploring data, and its role in the manufacturing industry. I'm your host Andrew Rieser. Today we are joined by Bastiane Huang. Bastiane is the product owner for OSARO. Welcome Bastiane.
Bastiane : Thank you so much, Andrew. It's my pleasure to be here.
Andrew: Likewise. So lots of really great topics, we really want to dive into today. But if you don't mind, just giving some brief background about yourself and also about us OSR.
Bastiane: Sure, I'm currently leading machine learning Basic robotic vision product strategy and development at OSARO. OSARO is a San Francisco-based startup building machine learning software for robotic vision and control, mostly in warehouses and factories. I personally have 10 years of experience in technology and automation, particularly in the area of artificial intelligence and the Internet of Things. Super passionate about machine learning, and its impact on human society, so to speak and write about machine learning and future-forward on Venture Beat, Harvard Business Review, my personal medium, how's that work. I also started an AI and MO patra manager Coverity in San Francisco to bring together hundreds of, machine learning professionals from companies like Facebook, Google, Amazon, Wei mo to share challenges and best practices of building machine learning products, because the area is so nascent. So we want to encourage more discussion, to talk about the unique challenges of building machine learning products and prior too so I, worked on product and business development for machine vision sensors and cameras, work for Amazon Alexa as a senior product manager and scale of EDA recognition software business from scratch. At Harvard Business School I want to switch from big companies to startups, to really focus on privatizing frontier technologies like machine learning. And that's why I reached out to Derik Pridmore was the CEO and co-founder of OSARO.
Andrew: Awesome, so a very diverse background for sure, with a lot of underlying themes around machine learning and artificial intelligence, which we'll dive into here in a little bit. So maybe you can explain a little bit about OSARO. So you mentioned it's a start up and you mentioned, it's in this space, and maybe you can kind of hit on some of the industry challenges that you're seeing, and why you feel like now's a great time for a company, like OSARO to start pioneering some of this information.
Bastiane: Sure. So right now, machine learning or deep learning, reinforcement learning are really buzzwords that used them in many different places. But OSARO is really uniquely positioned as a pioneer in deep learning and reinforcement learning, started working on deep learning and resource learning back in 2015, when the company was founded. So Derik was currently CEO, he was working for, peer to at Founders Fund, and he met with Deep Mind, Let the Cirrus be of Deep Mind. And after a year after Deep Mind got acquired, by Google in 2014. He went on to start his own company, which is OSARO. So since the beginning of the company been very focusing on productizing, deep learning, and reinforcement learning. So unlike a lot of other machine learning companies, we know we're not here to publish more papers, we'll do more research. We're really here to try to turn the latest research in machine learning into real-world products with actual value to customers. So we very much focused on products that can deliver true value, we specialize in AI defined robotics for factory and warehouse automation. And, our mission is to build machine learning software that enables robots to learn and adapt to changes in the environment.
Andrew: Yeah, that makes perfect sense. So essentially, it sounds like OSARO’s bringing together not only the technology but also the industry experts. And this productization that's occurring right now, is really what we're going to start diving into. So why is it important to bridge this gap between the technologies and expertise around this? And then more specifically, what are some of the applications that this is going to start providing value for in these industries?
Bastiane: Sure. So machine learning with deep learning has been around, for at least two decades, but it has recently become very popular because of the breakthrough we've seen over the past few years. But if you pay attention to the breakthrough, you'll see that it's mostly academic achievements, like being able to build a machinery system that can beat humans in NGO or video games, but those are not actual products that deliver real-world values. And to actually build real-world products, you need to be able to bridge these two groups, on the technologists and experts. So us at OSARO, we know a lot about technology, a lot about deep learning, meaningful learning, but those are transfer learning the latest research results. But we don't necessarily know what are different use cases, in warehouse in a factory or in restaurant industries. There are also a group of domain experts who know a lot about their individual film. And they're really interested in introducing machine learning to their individual industries, but they don't know how to start. So we really need more people who are able to speak, a language of both groups and make sure that they bridge the gap understand a lot about the user experiences and use cases in different industry but also understand to take knowledge, enough to know that if this is really physical, which use cases should we start first. So that's a lot of things I'm doing as a product manager at sorrow, but also a lot of things are also as the entire company, is working on. And to your second question about how machine learning enables robots to do different things. So robots have already been used in factories, manufacturing facilities, especially automotive, manufacturing facilities for many, many years, as a very mature industry, that most of the traditional robots are programmed to perform the same task over and over again, with high speed and high position. They can't really deal with any changes of surprises. So even if there's a small change in the process, switch to a new component or even just move the, component to the left by 10 centimeters. You need someone else to come in and reprogram the robot. You need engineering efforts to actually change process. You also need surrounding systems like feeders, conveyor belt or shaker table to make sure that components are delivered to a robust and the right place, right and go every time so the robot knows how to deal with it. So it's not a very flexible scenario. So when we're doing this, to use machine learning to allow robots to react to changes in the environment, learn to handle a wide range of different items have a range of different tasks. And more importantly, to learn, oh, this tasks was minimum human supervision. So this way, you can really save a lot of human costs you can really save on a lot of the surrounding system like the conveyor belt I just mentioned, these kinds of surrounding system are usually, more than four to five times of robot costs, so it's really significant. And lastly, it also enables robots to be used in new use cases. For example, you don't really see robot arms being used in warehouses right now. Because a typical warehouse that has millions of, different products is not feasible to program a robot, you're able to deal with a million different products in a million different ways. So now, because of machine learning robots can be used in this kind of less structured environments was much more varieties more complex tests. And we do see a lot of potential with this kind of machine learning enabled robotics in different industries.
Andrew: Yeah, I think one of the key points that you highlighted in there that always resonates with me. So when you think about manufacturing, bringing in automation, the goal is to improve processes increase quality, and kind of have that robot perform that those consistent tasks but then to your point, where that kind of gets derailed, is any changes in the environment or changes in the product, that's now coming down the conveyor belt, or things like that, that in essence requires reprogramming and either depending on external resources, because they may not have the skill sets internally at the manufacturing plant to make those changes. And so then when you factor all that in, the automation becomes less attractive to a certain extent. And so what I'm really intrigued by and I'm looking forward to conversing more about this is, there's some of these use cases and examples of where this, machine learning is going to essentially become the future, of how we start looking at automation within these different industries. So Bashing, Maybe you can talk to another common issue, that comes up the human workforce versus the robotic workplace and the dynamics around that, and how you all at a scenario are factoring that in and considering that?
Bastiane: Sure, in our view, this kind of AI-enabled robots are really designed to augment humans so workers can be freed up from repetitive or physically demanding tasks, and focus on tasks that require higher skills or deliver higher value. So take warehouse automation example we now have robotic picking system in three continents to do automated grocery fulfillment. So the robot can automatically recognize and pick dry food, or produce, even though each one might be different in terms of shape and size, but the robots can recognize a wide range of items with minimum human teaching or supervision. Right now if when you go to a lot of automated warehouses, you still see a lot of workers standing there, and for the entire day is basically taking items, from one box to the audit box, and it's really repetitive and boring and and basically pushing workers to their physical limits. That's why the turnover rate in warehouses is really high rate of 14%. And a lot of countries also facing a labor shortage. We have a lot of customers in countries like Japan and Germany. The aging population is really serious issue. So for example, Japan's population is expected to shrink by 24%, by the year 2025. And their working-age population, it's expected to shrink an even faster pace, than this overpopulation. And right now, the average age of Japan's farmer is already 70 years old. So it's not an issue that's happening in the future, is an issue that's happening right now. So that's why we started to help this customer Jojo made part of their tasks in their facilities step by step. And we also see more companies start to take action. Like for example, Amazon has always been a major driving force behind warehouse automation and same-day delivery. Last year, they announced a 700 million pledge returns to trend its workers into highly skilled jobs. And we think more companies will follow and do the same thing.
Andrew: Yeah, it's definitely interesting. I agree 100%, that this retraining is going to be needed, and that the automation and the robots, that we're referring to need to work in harmony with the humans on the shop floor as well. What's interesting that you guys are probably seeing, obviously with COVID and this pandemic that's going on. We're seeing a lot of manufacturers and just supply chain industry as a whole, accelerating their digital transformations and automation at such a ridiculous pace right now because this is the new reality of what's setting in. And so you see examples of factories having to shut down, because of a case of Coronavirus being discovered and potentially in the future. I think it'll be less dramatic impacts of things like this, if you're augmenting or building these factories of the future that are essentially built to withstand and withhold any of these real-world challenges, that often get presented.
Bastiane: Yes, for sure. I think we've seen a surge in demand on our customer side, especially because right now, a lot of them need to run automated warehouses 24 hours a day and seven days a week, and that definitely poses a challenge. For this kind of AI-enabled robotics, because this is still relatively new technology, compared to other kind of automation technologies, in factories, and warehouses. So, a lot of companies, they're still in pilot stage or proof of concept, but now certainly has to enter full production, and then no one can enter there to fix the machine, from time to time. I think on one hand, COVID-19 really drives companies to rethink their supply chain, and automation, but it also drives AI companies like us to focus more on building robust systems. So it's not just about performance metrics, like how fast can the robot run? How accurate can it be, but also, how many human intervention does it require per week? Because even if you have super fast, self-driving delivery robots If he has to stop every 15 minutes or every one hour, then it's not actually delivering a lot of value, to his customers. So I think we actually in a very interesting transition period, because from now on AI research is not just about AI dazzling demos for customers, he has to work in the real world. And he has to be very robust. So it's not just about machine learning. So it's about integration with the surrounding system. It's about software and hardware integration, as it was about user experience, error handling, if there's an error, how do you notify users? And how do you help users to do troubleshooting remotely? So we're really excited for this transition. And we also think there's a lot of surety, so there's actually a McKinsey report they rank industry by their potential to be automated. So they identified a view industry with high gifts, potential to be automated, their industries like accommodation and food services, warehouse logistics, and manufacturing. So in our opinion, warehouse automation is really just a starting point. It's kind of like a low hanging fruit, as the relatively more structured environment. And, and the tasks are simpler. But it's really just a start. As the technology improves in terms of his accuracy, in terms of his adaptability engine rally, we believe this kind of machine learning technology can be used in a lot more industries, like factories and food processing, etc.
Andrew: Yeah, that makes perfect sense. So maybe Can you describe a little bit more about some of the use cases around maybe OSARO’s vision and maybe the food or supply chain industry? Is there anything that you can kind of talk to you about a problem that you guys have encountered and then how you applied some of the specific technology solve that problem.
Bastiane: Sure. So in addition to the grocery fulfilling example I just talked about the warehouses. We also help our customers to identify a wide range of different components and warehouses and factories. So for example, we can help food packaging factory to identify 20 different transparent bottles, and then take the bottles from a jumbo bin, orient them and place it onto a packaging machine. Then the bottles can then be filled, labeled and kept later on in the packaging machine. So the most difficult part is really to be able to pick this kind of items from a jumble bin. If its already been separated on the conveyor belt, then it's actually very easy for machine vision cameras to identify you don't really need that much machine learning a traditional machine vision camera can do the job. So we're helping our customers to address the toughest challenge which to pick is from a jumbo bin. We also work was a restaurant customers to pick, sauce packets, chopsticks, and do those and then place them into a container in a bi-orientation direction. So this package of hot pads can be sent to their customers.
Andrew: Interesting. So before we sign off today, it would be great if you can share any last thoughts of where you see the future of this industry evolving and more specifically, where you guys are spending your time and energy as OSARO continues to evolve.
Bastiane: Sure, and seeing macro trends as the labor costs will continue to go up. And they do have halation is continue to be a more and more serious issue in many countries. And then we also see however it caused like robot cause a camera costs to continue to go down. So it makes more and more expensive for people to use automation and different countries, we started was countries with the highest labor costs like Japan or Germany, United States. But gradually we can expand our technology to other countries like Southeast Asia, China, etc. And in terms of industry, was started was warehouse information because it's easier use cases and warehouse, automation system integration there's, they're also very motivated to further automate their, process because they're facing a lot of competition from companies like Amazon mostly. So this is an industry we think is very good starting point and customers are familiar with robotic technology and they are willing to try different technology. There are other industries that have a lot of potential for automation, but customers are not necessarily aware of robotic or machine learning technologies like restaurants. through food processing, these are, these tend to be more traditional industries probably take more time for customer education to onboard new customers. But these are our future focus to in addition, to factory automation. So we're talking to a lot of fashi mission or system integrators, factories and also cosmetics companies, food companies to help them address different use cases.
Andrew: Very cool. Well, we really appreciate you spending the time to educate us a little bit more about not only OSARO, but where you see the the industry going. I definitely think that machine learning and deep learning, and automation, we have this perfect storm that's going onright now to where that education and the technology is catching up. And more importantly, some of these macro trends that you just described are taking effect. So should be some exciting times over the next decade as it relates to these technologies and look forward to reading up more and seeing a sorrows journey as you guys start tackling this more and more.
Bastiane: Thank you so much.
Andrew: For those that are listening if you'd like To learn more about SRO and their solutions, I'd encourage you to visit www.sro.com. That's OSARO.com. And if you'd like to connect with Bastion, we'll be sure to provide links to her information as well as the sorrows in the show notes. If you enjoyed this episode, please take a moment to rate the episode and subscribe to Data In Depth available on iTunes, Google, Spotify, Stitcher, and pretty much anywhere else you might listen to your podcasts. Thanks again for joining us today.
Announcer: Data in depth is produced by Mountain Point, a digital transformation consulting firm focusing on the manufacturing sector. You can find show notes additional episodes and more by visiting data in depth dot com. Thanks for listening and be sure to subscribe, wherever you get your podcasts.