Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Welding Robots Still Need Human Help
Transcript
- Lucas: You know that sharp, almost chemical smell you get near a body shop — that mix of burning metal and ozone? Luna: Yeah, I grew up near a plant that did chassis assembly. That smell sticks with you. Lucas: Right. And if you've ever watched someone weld — a human, with a hood and gloves — it's this incredibly precise, almost hypnotic dance. The welder is reading the puddle of molten metal, adjusting the angle, the speed, the heat, all in real time. Now, try getting a robot to do that. Luna: And I bet it's not easy. Because welding is one of those jobs we've been trying to automate for decades, and yet... a lot of critical welds are still done by people. Lucas: Exactly. And that's our topic today: why welding — a 3,000-year-old process — remains one of the hardest things to fully automate in manufacturing. I'm Lucas. Luna: And I'm Luna. This is The Robotics Podcast with Fexingo. Lucas: If today's tech conversation gave you something usable, something worth thinking about over coffee — we'd love it if you considered supporting the show. We deliberately don't run ads on these. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Luna: Totally. It's a small way to keep this conversation ad-free and independent. Now back to the weld — Lucas, what's the specific problem? Lucas: So the core issue is something called the 'weld gap.' When you're joining two pieces of metal — say, two sheets of steel that will form part of a car frame — they don't always fit perfectly. There's a tiny gap, maybe half a millimeter, maybe a millimeter. A human welder sees that gap and adjusts: they move the torch a little, they add a bit more filler material, they change the speed. A standard industrial robot, the kind you see on assembly lines, cannot do that. Luna: Because the robot is programmed for a perfect world — perfect fit, perfect geometry. And the real world isn't perfect. Lucas: Precisely. And it's not just the gap. When you weld, the metal heats up and expands. So the geometry actually changes while you're welding. A robot that's following a pre-programmed path will just keep going, even if the gap opens up or the metal shifts. And that leads to a weak weld, or a weld that fails inspection. Luna: So how big of a problem is this in practice? Like, what's the cost of a bad weld? Lucas: Huge. I talked to a quality engineer at an automotive supplier in Ohio — let's call the company Midwest Metalworks. They spent about $2 million trying to automate a single welding cell for a chassis component. The robot could do about 80 percent of the welds without issue. But that remaining 20 percent — the tricky ones with variable gaps — they had to have a human operator standing by to fix them. The robot would make a bad weld, the line would stop, and the human would come in with a grinder and re-weld it. Luna: So the 'lights-out factory' dream — fully automated, no humans — hits a wall with welding. Lucas: It does. And it's not just automotive. Shipbuilding, aerospace, even the construction of steel buildings — any place where you have thick metal, variable joints, and high-stakes welds. The economics of automating those last 20 percent are brutal. The sensors, the real-time control systems, the machine learning — it all adds up fast. Luna: But there are companies trying. I've heard of one called ArcVision. What are they doing? Lucas: ArcVision is a startup out of Pittsburgh. They're using thermal cameras — high-speed infrared — to watch the weld pool in real time. The camera captures the temperature distribution across the molten metal, and a machine learning model predicts whether the weld is going to be good or bad, about 200 milliseconds before a defect actually forms. Luna: So it's like giving the robot a sense of touch, but through vision. Lucas: Exactly. The system can then adjust the welding parameters — current, voltage, travel speed — on the fly. They've demonstrated it on some pretty gnarly joints, like T-joints with gaps that vary from zero to two millimeters. And they can get acceptable welds about 95 percent of the time. Luna: That's impressive. But 95 percent still means one in twenty welds might be bad. In a safety-critical application, that's not good enough. Lucas: You're right. And ArcVision is the first to admit that. Their CEO told me that for now, they see their system as a 'copilot' for a human welder — the robot does the easy welds, and the human handles the tricky ones or does post-weld inspection. They think fully autonomous welding is still about a decade away. Luna: So we're heading toward a hybrid model. Not humans vs. robots, but humans and robots working together in the same cell. Lucas: Yes. And that's actually the direction most factories are going. It's called 'collaborative welding' or 'human-robot collaboration.' The robot does the repetitive, high-speed welds on consistent joints. The human does the complex joints, the repairs, and the inspections. And the robot learns from the human — some systems even allow the human to manually guide the robot through a new weld path, which the robot then repeats perfectly. Luna: That sounds like a smart middle ground. But what about the cost? If a company has to buy both a robot and a human's time, does it still save money? Lucas: It depends on the volume and the complexity. For a high-volume production line — like a car door that's the same every time — full automation still wins. But for low-volume, high-mix manufacturing — say, a factory that makes custom machinery or ship components — the hybrid cell can be more cost-effective. You're not trying to replace the human; you're amplifying their productivity. Luna: That's a key distinction. It's not about eliminating jobs; it's about making the skilled welder more efficient. Lucas: Exactly. And the welders I've talked to actually prefer it. They say the robot takes care of the boring, repetitive welds that cause fatigue. They get to focus on the challenging, interesting work. And their value goes up — they become 'weld programmers' or 'robot operators,' which pays better. Luna: So the future of welding is not a robot alone, nor a human alone. It's a team. Lucas: That's the thesis. And I think it applies beyond welding — it's a model for how we should think about automation in general. The hard part isn't the technology; it's figuring out the right division of labor between human and machine. Luna: Before we wrap, what's one thing you learned that surprised you in this research? Lucas: I think the thing that struck me most is that the 'weld gap' problem is not just a technical challenge — it's an economic one. The cost of sensing and adapting in real time is still so high that it's often cheaper to have a human stand there and fix the errors. That tells me that the next big breakthrough in welding robotics won't be a better robot arm; it will be a cheaper, more robust sensor system. Luna: That's a great takeaway. And it leaves me wondering: will that sensor breakthrough come from a startup like ArcVision, or from a giant like Fanuc or ABB? Lucas: That's the million-dollar question — or maybe the two-million-dollar question, as Midwest Metalworks learned. I'd bet on a mix: startups pushing the frontier, and incumbents buying or licensing the tech once it's proven. Either way, the welder's job is changing. And that smell of burning metal? It's going to be around for a while longer. Luna: Good. Some things are worth keeping human. Thanks for listening, Lucas. Lucas: Thanks, Luna. And thanks to our listeners. If today's conversation sparked an idea, consider supporting the show at buy me a coffee dot com slash fexingo. We'll be back next week.