Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Hands Still Cant Hold a Hammer
Transcript
- Lucas: So there's a handful of tasks that have become the litmus test for robotic dexterity. Picking a screw from a bin, folding a towel, threading a needle. Today I want to talk about one that's even more brutal: hammering a nail. Luna: Hammering? That seems almost primitive compared to the other ones. You'd think a robot could just swing hard and hit the nail. Lucas: Right, and that's exactly what most robots do if you let them loose. They swing with massive force, and the nail either bends, the wood splits, or the hammer bounces right back. The problem isn't raw power. It's controlling what happens in the milliseconds after impact. Luna: So it's not about strength, it's about the rebound. That actually makes sense. When you hammer, you're fighting the hammer bouncing back. Lucas: Exactly. And that's why hammering is such a great test case. It forces a robot to deal with dynamic contact, which is completely different from the quasi-static tasks like picking or placing. When a hammer hits a nail, there's a sudden exchange of energy. The hammer decelerates from maybe five meters per second to zero in a few milliseconds, and that creates forces that are way beyond what a typical force sensor can measure quickly enough. Luna: So the robot's brain can't even perceive what's happening fast enough to react. It's like trying to catch a ball that's moving at the speed of sound. Lucas: That's a great analogy. And it's not just about speed. The hammer's mass and the stiffness of the grip change the whole dynamics. If you grip the hammer too rigidly, the impact force goes directly into the robot's arm, which can damage the gears and joints. If you grip too loosely, the hammer slips or twists in the hand. Luna: So you need a grip that's compliant but not too compliant. There's a sweet spot. Lucas: Right. And that's where impedance control comes in. Instead of trying to control position or force directly, you control the relationship between them, like a spring with adjustable stiffness and damping. With the right impedance, the robot can absorb the impact, maintain control, and still deliver enough energy to drive the nail. Luna: But that sounds like a lot of tuning. Is that something you can just do on the fly? Lucas: For now, it's mostly pre-computed for a specific hammer and nail. But researchers are starting to use machine learning to adapt the impedance parameters based on the sound and the vibration feedback. They're basically letting the robot learn the feel of a good hit versus a bad one. Luna: So the robot is listening to the impact and adjusting its stiffness and damping for the next swing. That's like a human feeling the hammer rebound and adjusting their grip. Lucas: Exactly. And it's a perfect example of why manipulation is so hard. We don't think about it, but every time we hammer, we're processing a huge amount of sensory information in real time to modulate our muscles. The robot has to do the same, but with sensors that are often slower and less rich than our skin and muscles. Luna: It's amazing we ever learned to do it ourselves. And I guess that's why robots are still stuck at the screwdriver stage. Lucas: Yeah, and that's a good segue. If today's conversation gave you something useful—maybe a new appreciation for why hammering is a brutal test of robotics—we'd love to keep doing these deep dives. The way we keep this show ad-free, actually the way the entire Fexingo network stays ad-free, is listener support. If you want to help, you can buy us a coffee at buy me a coffee dot com slash fexingo. Luna: And it genuinely helps us spend more time on research and less on sponsors. We really do read those messages too. Lucas: So back to the hammer. The interesting thing is that most robotic manipulation research has focused on non-percussive tasks, like grasping and placing. Hammering is in a different category entirely, and it's only recently that labs have started to seriously tackle it. Luna: What's the actual state of the art? Can any robot hammer a nail into a wall and not destroy it? Lucas: As of August 2026, we're not quite there. There are a few research demos—like the one a few months ago where a robot drove a nail into a pine board using a standard claw hammer, but it took about thirty seconds per nail and it needed a camera to locate the nailhead. And it was a controlled setup. No angle changes, no wall studs, no surprises. Luna: So it's like the first time a robot folded a towel. It works, but it's not ready for your kitchen. Lucas: Exactly. And the harder part is when the nail is angled or the surface is different. You don't just swing harder; you adjust your strategy. Sometimes you need a few light taps to set the nail, then solid blows. And if the nail bends, you have to pull it out and start over. That's a whole other manipulation skill—prying, which is also percussive. Luna: So it's not just about the hammering, it's about the whole job. And that's why I think construction robots are still mostly limited to bricklaying or painting, where you don't have that dynamic contact. Lucas: Right. And there's a startup I was reading about that's trying to tackle just the hammering step, but with a specialized tool that's more like a nail gun than a hammer. They're arguing that you don't need a humanoid to hammer; you need a purpose-built end effector that can drive fasteners. But the challenge is that nails are cheap and everywhere, and a nail gun requires compressed air or a cartridge. Luna: So the robot might be better off using a nail gun, but then you're adding complexity and cost. And you're also not solving the general problem of dynamic manipulation. Lucas: That's the tension. But the research on hammering is pushing the boundaries of what we know about controlling impact. And it has implications beyond construction. Think about any task that involves a sudden force: chiseling, cutting with a knife, breaking an egg. All of those require the same kind of dynamic control. Luna: And we've done an episode on breaking an egg, so we know that's hard. But hammering is even more percussive, I guess. Lucas: It's a good point. And that's why I think hammering is the next big frontier. We've mastered the slow, precise stuff. Now we're moving into the fast, messy stuff. Luna: So what's the one takeaway you want listeners to have today? Lucas: I think it's that the hardest part of robotic manipulation isn't precision or strength—it's dealing with dynamic contact. And hammering is the perfect example. It's a simple tool, a simple task, but it requires a level of real-time control that we're only starting to achieve. Luna: And next time you hammer a nail, you can think about the physics and the milliseconds of impact. I know I will. Lucas: That's the spirit. And if you have a topic you want us to tackle, we're all ears. But for now, thanks for thinking with us.