Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Handle a Wet Bar of Soap
Transcript
- Lucas: Alright, Luna, I have a question for you. How do you pick up a wet bar of soap? Luna: Carefully? With two hands, maybe a towel. Why? Lucas: Because that is still a genuinely hard problem in robotics. We've done fifty-something episodes on why robot arms can't handle eggs, tomatoes, spaghetti, rice grains, human hair. But wet soap — it's almost the perfect storm of failure modes. It's slippery, deformable, and if you squeeze too hard it shoots out of your grip. And unlike a tomato, you can't just pierce it. Luna: So this is the next frontier in dexterous manipulation. What's the current state of the art? Lucas: Well, a team at MIT's CSAIL published a paper earlier this year — they built a system that uses tactile sensing and real-time machine learning to predict slip. They tested it on wet soap, among other things. With a standard parallel-jaw gripper, their success rate went from 12 percent to 73 percent. That's a huge jump, but it's still not reliable enough for any production environment. Luna: Seventy-three percent means one out of four times it still fumbles. That's not good enough for a factory line packing soap. Lucas: Exactly. And that's the core tension. Humans have about four million nerve endings in our fingertips. We have this incredible ability to modulate grip force subconsciously. A robot with a standard gripper has zero tactile feedback unless you add sensors. And even then, the physics of wet contact is brutal. Luna: Let's talk about the physics. Why is wet soap so much harder than, say, a dry block of wood? Lucas: It comes down to friction. Most robotic grippers rely on Coulomb friction — the dry contact between two surfaces. The coefficient of friction determines how much force you can apply tangentially before slip occurs. Water on soap acts as a lubricant, drastically reducing that coefficient. So the gripper has to increase normal force to compensate. But soap is also deformable, so if you press too hard, the contact area changes, the soap squishes, and it can pop out. There's a whole nonlinear dynamics problem. Luna: So it's not just a sensor problem, it's a control problem. You need to know the friction in real time and adjust force accordingly. Lucas: Right. And the MIT team used a tactile sensor called Gelsight — it's a gel-based optical sensor that captures high-resolution deformation of the contact surface. They trained a neural network to predict the slip margin — basically, how close the grip is to slipping — from the tactile images. Then they used that to modulate the grip force in a closed loop. That's how they got from 12 to 73 percent. Luna: That's impressive, but still not production-ready. Are there other approaches? Lucas: Some labs are exploring biomimetic fingertips with micro-scale ridges that mimic human fingerprints. The idea is that the ridges channel away fluid and increase friction in wet conditions. There's also work on electroadhesion — using electrostatic forces to grip, which works even on wet surfaces. But those are still very experimental. Luna: And I imagine the industry is watching closely. If you can reliably handle wet soap, you can probably handle a lot of other slippery items in food processing or pharmaceuticals. Lucas: Exactly. The same technology could be used for picking up raw chicken, wet lettuce, or medical vials with condensation. It's a benchmark problem. So solving wet soap unlocks a whole class of manipulation tasks. Luna: We should note that this research is part of a broader push. The National Robotics Initiative in the US has funded several projects on dexterous manipulation with tactile sensing. And companies like Soft Robotics have commercial grippers that handle delicate items, but they still struggle with wet, slippery ones. Lucas: Right. Soft Robotics uses inflatable rubber fingers that conform to objects, which helps with deformable items. But on wet soap, the rubber itself gets slippery. It's a tough materials science challenge too. Lucas: You know, this kind of deep-dive into a single hard problem is exactly why we do this show. And it's only possible because of our listeners. A handful of you chip in monthly at buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these episodes — no ads, no sponsors. So if today's tech conversation gave you something usable, that's a great way to keep it coming. Luna: Yeah, it really does make a difference. We appreciate every bit of support. Lucas: So back to the soap. Another challenge is that soap changes shape as you grip it. It's viscoelastic — it flows and deforms over time. So the contact patch isn't static. The robot has to constantly re-estimate the geometry. Luna: Which is computationally expensive. How are they handling that in real time? Lucas: The MIT system runs the slip prediction model at about 60 hertz — 60 times per second. That's fast enough to adjust force before the object slips. But the model was trained on a specific soap shape — a standard rectangular bar. If you give it a different shape, like a round soap, the performance drops. Generalization is still an open problem. Luna: So we're not at the point where a robot can handle any wet object. It's still very task-specific. Lucas: Exactly. And that's the story across most of manipulation. We've had industrial robot arms for decades, but they operate in highly structured environments. The moment you introduce variability — different shapes, surface textures, wetness — they fall apart. Wet soap is just a particularly vivid example. Luna: Is there any commercial application that's close to deployment? Like, are we going to see a robot that can pick up a wet soap in a factory soon? Lucas: Maybe not for soap specifically, but for other tasks. There's a company called RightHand Robotics that uses tactile sensing for warehouse picking. They handle a lot of different items, but they still struggle with slippery or flimsy packaging. So the research from MIT could feed into their systems. But I'd say we're probably three to five years away from a reliable commercial solution for wet, slippery objects. Luna: That's a realistic timeline. And it's worth noting that even with all this progress, a human can pick up a wet bar of soap with near 100 percent reliability, without thinking. That's the bar. Lucas: Right. And that's what makes robotics so fascinating. We take for granted the incredible sensorimotor control that evolution gave us. Replicating that in hardware and software is one of the grand challenges of engineering. Luna: What other objects are on the 'hard' list that we haven't covered yet? Lucas: Oh, plenty. A peeled hard-boiled egg — it's slippery and deformable. A live crab — it moves unpredictably. A piece of plastic wrap — it's thin and sticky. But wet soap is a great representative because it combines so many failure modes. That's why it's become a benchmark in the manipulation community. Luna: And if you solve it, you likely solve a whole class of problems. That's the hope. Lucas: Exactly. So next time you're in the shower and you drop the soap, just remember — it's not you, it's the physics. Luna: And the robots have it even worse. Thanks, Lucas. Lucas: Thanks, Luna.