Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Thread a Needle
Transcript
- Lucas: So today we are picking up where we left off with tasks that look trivial to us but are still a massive challenge for robot arms. And I want to talk about threading a needle. Luna: Oh, that is a good one. Because it sounds so simple — you just push a thread through a tiny hole. But for a robot, that's a nightmare. Lucas: Exactly. And we've done episodes on picking up a single grape, folding a fitted sheet — all these tasks that require a combination of precision, force control, and adaptability. Threading a needle is the same kind of challenge but with even tighter tolerances. Luna: How small are we talking? Like, the eye of a typical sewing needle? Lucas: A standard needle eye is about 0.5 millimeters wide. The thread is around 0.2 millimeters in diameter. So you have a 0.15 millimeter margin of error on each side. That is about the width of a human hair. Luna: Right, and we already covered how robot arms struggle with a human hair — that was episode 65. So threading a needle is essentially that same precision challenge, but now you're also trying to push something through a hole. Lucas: Exactly. And the thread is flexible, which makes it even harder. A rigid needle — like a hypodermic — is one thing. But a soft, bendy thread? That's a whole different ballgame. Luna: So what are the specific failure points? I imagine vision is a big one. Lucas: Vision is the first hurdle. The robot needs to locate the needle eye with sub-millimeter accuracy. Most industrial robot arms have a repeatability of around 0.02 to 0.1 millimeters — so in theory, they can position a tool with enough precision. But the vision system has to identify the exact position and orientation of the eye, and that can be tricky because of lighting, shadows, the shiny metal surface. Luna: And the eye is usually not perfectly round — it's often an oval or a slit, depending on the needle type. Lucas: Right. So you need a camera with high resolution and good lighting, and then you need to calibrate the robot's coordinate system to the camera's so that when the camera says 'the eye is here,' the robot actually moves to that exact spot. Calibration drifts over time, especially if the robot has been moving a lot or if the temperature changes. Luna: So even if the robot can theoretically position to within 0.02 millimeters, the whole system — vision, calibration, control — might only be accurate to 0.5 millimeters or worse. Lucas: Exactly. And that's before we even talk about the thread. The thread is soft and it bends. So even if the robot brings the tip of the thread to exactly the right spot, the thread might buckle or deform against the needle, or it might miss by a tiny amount and slide along the surface. Luna: That's where force control comes in. The robot needs to feel when the thread is touching the needle and then adjust. That's something humans do without thinking — you feel the resistance and you back off or change the angle. Lucas: Right. Force control is an active area of research. Most industrial robots don't have sensitive force feedback built in. They have torque sensors at the joints, but they're not precise enough for this kind of delicate contact. You'd need a force-torque sensor at the wrist, which adds cost and complexity. Luna: And then there's the issue of the thread itself. It's not just flexible — it's also frayed at the tip, or it might have a slight twist. That changes how it interacts with the needle. Lucas: Yes. A human can trim the thread or lick it to make the tip stiff. A robot could theoretically do the same — there are systems that use a small blade to cut the thread at an angle to create a sharp tip. But that adds another step and another potential failure mode. Luna: So what's the state of the art? Are there any robots that can reliably thread a needle? Lucas: In controlled lab settings, yes. Researchers have demonstrated needle threading with specialized robotic systems. For example, in 2021, a team at the University of California, Berkeley used a two-armed robot with high-resolution cameras and force sensors to thread a needle with about a 90 percent success rate. But that was in a carefully calibrated environment with good lighting and precise positioning. Luna: So not something you'd see on a factory floor. Lucas: Not yet. And the applications where it would be most useful — like surgical suturing or textile manufacturing — have other constraints. In surgery, the needle is often curved, and the thread is different, and the environment is wet and messy. In textiles, you're dealing with high-speed machines that need to thread needles thousands of times per day. Luna: I read that some modern sewing machines actually have automatic needle threaders — but they're mechanical, not robotic. They use a little hook that grabs the thread and pulls it through. Lucas: Exactly. Those are purpose-built mechanical solutions, not general-purpose robot arms. They work because the needle and thread are in a fixed position relative to the machine. A robot arm that has to pick up a needle from a tray, orient it, and then thread it — that's a much harder problem. Luna: So it's a good example of how far we still are from human-level dexterity. And it's not just about precision — it's about adaptability. Humans can thread a needle even if the lighting is bad, or if the needle is slightly different, or if their hands are shaking a little. Lucas: That's the key insight. The problem isn't just about hitting a small target — it's about handling uncertainty and variability. And that's what makes it such a rich research area. Luna: Honestly, if today's episode gave you something usable — a new angle on precision, or just a good conversation starter — you know where to find us. It's buy me a coffee dot com slash fexingo. Small ask, big help. Lucas: Yeah, that link really does keep the lights on. And we appreciate it. So — back to the needle. One area where this is really pushing forward is surgical robotics. Luna: Right. In surgery, you're often suturing with a needle that's already attached to a thread. The robot has to position the needle at the right angle and push it through tissue, then pass it to the other hand. That's a form of needle manipulation, but it's not threading a separate thread through a separate eye. Lucas: True. But the same underlying skills are needed: precise positioning, force control, and handling flexible materials. The da Vinci surgical system, for example, uses wristed instruments that mimic human hand movements. But even da Vinci doesn't thread a needle from scratch — the needle comes pre-threaded. Luna: So the frontier is really about combining vision, force, and control in a way that's robust enough for unstructured environments. And threading a needle is a perfect benchmark. Lucas: Exactly. And that's why it keeps showing up in robotics competitions and research papers. It's a microcosm of the whole manipulation problem. Luna: I want to ask about the thread itself — are there any clever tricks that robots use to make it easier? Like, do they use a stiff thread or a special coating? Lucas: Some researchers have used a technique called 'thread stiffening' — they apply a small amount of adhesive or wax to the tip of the thread to make it rigid. But that adds complexity and might not be suitable for all applications. Another approach is to use a vacuum or a tiny gripper to hold the thread absolutely straight. Luna: So there are workarounds, but they're not general solutions. That's the story of a lot of robotics problems — you can solve it in a specific case, but the general case is still hard. Lucas: Right. And that's what makes it interesting. The needle-threading problem touches on perception, control, mechanics, and materials science. It's a beautiful example of the kind of integrated challenge that robotics researchers love. Luna: And the kind of thing that makes you appreciate how amazing your own fingers and eyes really are. Lucas: Absolutely. So next time you're sitting there grumbling about a stubborn thread, just remember — a robot would probably have already given up and thrown the needle across the room.