Dark Factories: What Robots Actually Can (and Can't) Do | Azumuta Podcast Ep. 1

Will the factory of the future really run in the dark, with no humans in sight? In Episode 1 of the Azumuta podcast, two robotics and AI researchers separate the hype around dark factories from what the technology can actually do today. Dark factories (also called lights-out manufacturing) are one of the most discussed and most misunderstood ideas in modern manufacturing. So we went to the source. In this first episode of a four-part series, we sit down with Professor Francis Werfels and researcher Andreas Verlaenen, whose lab won an international robot cloth-folding competition, to look at the technological reality behind the headlines. This conversation is built for manufacturing leaders, plant managers, and operations teams who want signal instead of noise. We dig into where robotics, AI, and automation actually stand, and why "dark" doesn't have to mean "no people." In this episode you'll learn: 🔹 How the definition of a dark factory has shifted from high-volume, low-mix automation to high-mix, low-volume flexibility 🔹 Why robots still struggle with generalization, and what the famous towel-folding test really proves 🔹 The reason humanoid robots and collaborative robots still need clear work instructions to perform a task 🔹 What genuinely changed with vision-language models, large behavior models, affordable robotic hands, and tactile sensing 🔹 Why the robotics field may be "100 years" away from a ChatGPT-sized training dataset at current data-collection rates 🔹 An honest comparison of Optimus, Figure, Unitree, Boston Dynamics, and Toyota's research approach 🔹 Why small and mid-sized manufacturers should treat their data as a competitive moat, and start collecting it now The throughline matches how we see Industry 4.0 and Industry 5.0 unfolding: machines automate, but people stay in control. The factories that win are the ones that capture expertise as data and standardize how work gets done, so both humans and robots can execute it reliably. ⏱️ Timestamps 00:00 – Welcome: meet the cloth-folding robotics lab 02:00 – Has the definition of a "dark factory" changed? 04:00 – The towel test: what robots can and can't do 07:30 – Why robots still need work instructions 08:30 – What actually changed: hands, vision-language models, common sense 13:00 – The data problem: "100 years to match ChatGPT" 17:00 – Touch, teleoperation, and human-robot collaboration 20:00 – Optimus, Figure, Boston Dynamics, and Toyota compared 25:00 – When will you have a robot helper at home? 29:00 – Open source vs. closed source robotics 32:00 – Do large manufacturers have a data moat? 35:00 – Should you start collecting factory data now? 36:30 – How to collaborate with the lab 🔗 Resources & links 🌐 Digitize and standardize your shop floor: https://www.azumuta.com 📅 Book a demo: https://www.azumuta.com/book-a-demo 📘 Digital work instructions: https://www.azumuta.com/digital-work-... 💬 Follow Azumuta on LinkedIn for more manufacturing insights Enjoyed the conversation? 👍 Like the video, and tell us in the comments: do you believe in fully lights-out factories, or will humans always stay in the loop? Subscribe so you don't miss Episodes 2 to 4, where we hear from business leaders, experts, and labor unions. Share this with a colleague who's making automation decisions this year. #DarkFactory #DarkFactories #LightsOutManufacturing #SmartFactory #Manufacturing #Robotics #HumanoidRobots #Industry40 #Industry50 #FactoryAutomation #ArtificialIntelligence #CollaborativeRobots #ManufacturingTechnology #Azumuta #FutureOfManufacturing