Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Handle a Power Cord
Transcript
- Lucas: Okay, Luna, I want to start with something that sounds trivial but is actually a nightmare for robotics engineers: plugging in a power cord. Luna: A power cord. Like, the three-pronged kind we stick into a wall outlet? Lucas: Exactly. The kind that every human does in under a second, often without looking. Last month, a team at MIT CSAIL published a paper on this exact task. They found that a state of the art robot arm takes about two seconds — two thousand milliseconds — to successfully insert a plug into a socket. A human takes about three hundred milliseconds. Luna: That's a factor of about six and a half. But why is it so much slower? It's not like the robot has to hold a conversation while doing it. Lucas: Right. The paper breaks it down into three specific failure modes. First, visual occlusion. The moment the plug approaches the socket, the robot's own gripper blocks the camera's view of the socket holes. So the robot is effectively blind at the most critical part of the motion. Luna: So it's feeling its way in, not seeing. And robot joints are stiff — they don't have the compliance of a human wrist. Lucas: That's failure mode two. Human wrists have a natural elasticity. You can feel the plug align itself, and your wrist gives a little to correct the angle. A typical industrial robot arm has a stiffness of around one to two thousand Newtons per millimeter. That's like trying to plug something in with a crowbar. If the plug is off by even a millimeter, the robot either jams it against the socket or misses entirely. Luna: And failure mode three? The cable itself? Lucas: Yes. The power cord is a flexible, bendable, unpredictable object. As the arm moves, the cable drags on the table, twists, and applies unknown forces to the plug. The robot has no model of how the cable is going to behave. So it's trying to solve a physics problem with too many variables. Luna: So you've got blindness, stiffness, and a floppy tail. That's a tough combination. What's the leading approach to fix it? Lucas: The MIT team used a deep-learning-based tactile sensor on the gripper's fingers. It's essentially a high-resolution pressure map — like a fingerprint reader but sensitive enough to detect the subtle tilt of the plug. They trained a neural network on thousands of simulated insertion attempts, then fine-tuned on a real robot. Luna: Did it close the gap? Does the robot now take, say, four hundred milliseconds? Lucas: Not yet. With tactile feedback, the robot's success rate went from about forty percent to seventy percent in lab conditions. But the time dropped only to about one point five seconds. Still five times slower than a human. And in unstructured environments — like a dimly lit room with a tangled cable — it fell back to sixty percent success. Luna: So we're not plugging in our own vacuum robots anytime soon. But this isn't just about convenience, right? There are applications where plugging in is safety-critical. Lucas: Absolutely. Think about surgical robots in an operating room. They need to be plugged in to charge or to connect to instruments. If a nurse has to manually plug and unplug, that's a contamination risk. Or consider home-service robots that are supposed to dock themselves. Every major robotics company — from Boston Dynamics to the household robot startups — has struggled with the docking problem because it's fundamentally an insertion task. Luna: And the docking problem is essentially plugging in, but with a bigger connector. I've seen videos of robots bumping into charging stations for minutes before connecting. Lucas: Exactly. The charging dock is a low-tolerance version of a wall socket. And the same failure modes apply: visual occlusion from the robot's own body, force mismatch, cable drag. So solving the power cord problem would unlock a whole class of autonomous behaviors. Luna: What about using a softer, more compliant arm? Like the kind with series elastic actuators? Lucas: That helps with the stiffness issue. Some research groups have used variable impedance control — where the robot's joints soften on contact. But then you trade off precision: a compliant arm can wobble when trying to align the plug. It's a control-loop challenge. The MIT paper actually combined a stiff arm with a soft, sensorized fingertip. They got the best of both worlds. Luna: So the fingertip acts like a human finger pad — it deforms slightly and measures that deformation. That's clever. But it still needs the visual occlusion solved. Lucas: One idea is to use a second camera mounted on the wrist, looking at the socket from a different angle. But that adds cost and calibration complexity. Another approach is to use a force-torque sensor at the wrist to measure the reaction forces as the plug touches the socket face, then use that signal to guide the insertion. The problem is that the signal is very noisy — especially when the cable is tugging. Luna: So we're back to the cable being the wild card. If you could model the cable's behavior in real time, that might be a breakthrough. Lucas: Right. Some researchers are using what's called a 'cosserat rod' model — it's a mathematical model for flexible rods. It can simulate the cable's bending and twisting. But solving it in real time is computationally expensive. A typical simulation takes about ten milliseconds per timestep. For a control loop running at a kilohertz, that's too slow. Luna: So the cable is the unsung villain of manipulation. I bet the next big advance will come from a combination of tactile sensing and a learned model of the cable's dynamics — maybe a neural network that predicts the cable's behavior from past motion. Lucas: That's exactly where the field is heading. And it's why this problem is so fascinating: it forces you to integrate vision, touch, and physics simulation into a single control policy. It's not just about plugging in a cord — it's about how robots handle uncertainty in the physical world. Luna: It's also a reminder that the things we find easiest — like plugging in a lamp — are often the hardest for machines. And that's what keeps this field interesting. Lucas: You know, that kind of deep-dive conversation is exactly why we keep this show ad-free. We don't have to pander to sponsors or keep episodes short for ad breaks. If you find value in that approach, and you want to support it, you can find us at buy me a coffee dot com slash fexingo. No pressure — just a way to keep the content independent. Luna: Yeah, and it's listeners like you who make that possible. It's a small gesture that goes a long way for a niche show like this. Lucas: So back to the cord. The MIT team's next step is to integrate their tactile sensor with a wrist-mounted camera and a learned cable model. They're aiming for a demo later this year where the robot plugs in a standard NEMA five-fifteen plug — the one you see in every home — with ninety-five percent reliability. Luna: If they hit that, I think we'll see it in commercial service robots within a few years. And then maybe my Roomba won't need me to line it up with the dock anymore. Lucas: That's the dream. But for now, every time you plug in your phone charger, take a moment to appreciate the incredible sensorimotor coordination it takes. Your brain is solving a problem that the world's best robots still struggle with. Luna: And your wrist is way more compliant than a crowbar. Thanks, evolution. Lucas: That's it for this episode. We'll be back next week with another look at the small motions that trip up big machines.