How AI is Transforming Manufacturing End-to-End

I think that what's happening now with the use of AI, the way that we're using AI in our everyday lives, as a way to augment our own capabilities, it's something that's going to come into manufacturing much, much faster than anything previous to it. We're in the process of fundamentally reimagining the design, build, and operate phases of the lifecycle of products. What AI is doing is helping us understand how we collapse a very complex environment down into a very simple one loop for engineering, design, manufacturing together. Just imagine the design phase of an industrial product. Today, you've got to wait for the actual product to be fully designed and ready before you start industrializing that product for production. Think of Foxconn. In this case, ABB are looking at a consumer electronics device. There are some very, very small parts in that device that a typical camera can't see accurately. So being able to then just look at a certain position and then use synthetic data to understand where the robot can accurately grip that part and manipulate it produces a significant improvement in productivity. Our collaboration with NVIDIA, together with ABB's own robot studio capability, has now meant we've managed to fully integrate the complete stack and optimize it to a point where we're now achieving 99% accuracy on the simulated version to the real version. When it comes to AI trust, what we need is clear boundaries and clear principles about how we manage our data. And we need to always be able to refer to the ground truth of physics simulation, alongside the surrogate models. What we're doing is we are working with Neural Concept on a design lab. So what that means is it's an AI-first platform and it leverages surrogate models in aerodynamics, allows us to make decisions in real time in one place, so design and engineering in one loop. We have a tool called Tulip Factory Playback, and this is a way to bring video streams from what's happening on the floor, transactional information that's coming from Tulip, sensor data that's coming from machines, and bring that together into a profiler or debugger for your factory floor. We're using, from the Metropolis platfom, video search and summarization blueprints. So what you can do with the language models and with the vision language models now is interpret that human behavior in a way that you couldn't before. We've deployed Factory Playback at Terex, a maker of industrial machineries. Every one is designed and customized to order. So the optimization of that line becomes incredibly challenging. If you can make a 1% difference in the operational efficiency of the whole line, that's immediately millions of dollars of ROI for them. Now that we've solved the sim-to-real gap, we can now really integrate physical AI much more elegantly into the robot systems. So I see the future very much as we're able to understand how do we make our products better, because we have more data in one place with the expertise in the loop. I'm just excited to see what people are going to do when they have that power on the manufacturing side.