Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Can't Insert a USB Plug
Transcript
- Lucas: So here's a test you can run at home. Pick up a USB cable and plug it into your laptop. You didn't even think about it, right? Your hand just found the orientation, felt the resistance, and pushed. Luna: And if you get it wrong the first time, you flip it and it works. Takes maybe two seconds. Lucas: Now ask a robot arm to do the same thing. It's a disaster. Even in 2026, most industrial robots fail to insert a USB plug on the first try about 30 percent of the time. And when they do succeed, it takes them three to five seconds of fumbling. Luna: That's wild. I mean, we've sent robots to Mars. We have autonomous cars. But plugging in a cable is still hard? Lucas: It's genuinely hard. And it's one of those problems that sounds trivial until you break down what's actually happening. A USB plug is symmetrical in only one orientation—you have to align a specific shape into a specific hole. The clearance between the plug and the receptacle is about 0.4 millimeters. That's tight. Luna: And the robot can't see the alignment perfectly because its camera might be occluded by its own gripper, or the lighting creates shadows on the plug. Lucas: Exactly. So the robot relies on force feedback—sensors in the wrist or fingers that measure resistance as it pushes. But here's the problem: if the plug is misaligned by even a fraction of a degree, the forces you feel are nonlinear. The plug jams, the robot pushes harder, and now you've got a bent pin or a broken connector. Luna: I remember reading a paper from MIT CSAIL a couple years ago—they built a system that used a search strategy. Instead of trying to align perfectly, the robot would wiggle the plug in a small spiral while pushing gently. Lucas: Right, that's the classic approach. The robot doesn't try to know the exact position; it searches for the hole. That paper achieved about 95 percent success in lab conditions. But the search time was variable—sometimes half a second, sometimes three seconds. And it only worked with rigid connectors, not the slightly flexible cables you find on consumer electronics. Luna: So the cable itself introduces another degree of freedom. The plug isn't rigidly attached to the robot—the cable bends and twists, adding uncertainty. Lucas: Exactly. And that's the gap between lab and factory floor. In manufacturing, you see this problem with harness assembly—robots trying to insert dozens of connectors into a backplane. It's still mostly done by human hands because the cost of a single failed insertion—a scratched pin, a broken connector—is higher than the labor cost. Luna: There's a company called Assembly Robotics that tried to solve this with a vision-guided system. They used multiple cameras and machine learning to estimate pose. I think they got to about 98 percent success, but the cycle time was too slow for high-volume production. Lucas: And that's the trade-off. You can brute-force the problem with better sensors and more compute, but now your robot cell costs 200 thousand dollars and takes three seconds per plug. Meanwhile, a human does it in half a second with zero capital cost. Luna: So where is the research heading now? Are we ever going to see a robot that can plug in a USB cable as reliably as a person? Lucas: I think the real breakthrough will come from tactile sensing that mimics human skin. There's work at UC Berkeley on soft fingertips with embedded microfluidic channels—they can detect shear forces and slippage in real time. Combine that with a learned compliance model, and you might get human-level performance. Luna: But that's still in the lab, right? No commercial robot has that kind of tactile resolution yet. Lucas: Not yet. The best commercial force-torque sensors have a resolution of about 0.1 Newtons. Human fingertips can detect forces as small as 0.01 Newtons. We're an order of magnitude off. And that gap explains why your robot vacuum cleaner still gets stuck on a rug, and why a robot can't plug in a phone charger. Luna: It's humbling. We take for granted how much our sense of touch is doing in the background. Lucas: Speaking of which—if this kind of deep-dive into robotics is something you find valuable, it's worth noting that this show is entirely listener-supported. We don't run ads. And it's listeners like you who make it possible for us to spend time on topics like why USB insertion is still hard. If you'd like to help, you can find us at buy me a coffee dot com slash fexingo. Luna: Yeah, every contribution, no matter how small, goes directly into research time and production. It keeps the podcast ad-free and focused on the real engineering challenges. Lucas: Alright, back to the tech. So one promising approach that's getting traction is what researchers call 'learned compliance.' Instead of programming the robot's force response analytically, you train a neural network on thousands of insertion attempts—both successes and failures—so the robot learns the feel of a correct insertion. Luna: That's essentially what OpenAI did with their Dactyl hand for cube reorientation. But that required millions of simulated attempts. Lucas: Right. And simulation is cheap—you can run a million USB insertions in a day. But the sim to real gap is brutal. The friction, the cable flexibility, the slight deformations of the plastic—all of that is hard to model accurately. So the robot that succeeds in simulation often fails in the real world. Luna: There's also the standardization question. usb c was supposed to make this easier, and it does—because it's reversible, you don't need to orient the plug. But the insertion force is still about 10 to 15 Newtons, and the alignment tolerance is similar. Lucas: So even reversible connectors don't eliminate the need for fine force control. They just reduce the orientation search from two possibilities to one. The fundamental problem remains: you have to insert a peg into a hole with very little clearance, without damaging either part. Luna: And that's the same problem as inserting a battery into a smartphone, or placing a chip into a socket, or connecting a hydraulic line. It's everywhere in manufacturing. Lucas: Exactly. That's why this seemingly trivial task is a benchmark for the entire field of robotic manipulation. If you can solve USB insertion reliably, you can solve a huge class of assembly problems. And right now, the solution is still on the horizon. Luna: So next time you plug in a cable without thinking, maybe take a second to appreciate what your nervous system is doing. It's one of the hardest tricks in robotics. Lucas: And maybe spare a thought for the robot arm in the lab, still trying to get it right.