Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Handle a Sponge
Transcript
- Lucas: Alright Luna, quick question: how much force does it take to pick up a wet sponge? Luna: I... I don't know. Almost nothing? You just grab it. Lucas: Right. For a human it's trivial. But for an industrial robot arm — and we're talking arms that can lift a hundred pounds with repeatability of a few microns — a wet sponge is a nightmare. And that's what we're digging into today: why a three-dollar kitchen sponge is still, in the year 2026, a genuine unsolved problem in robotics. Luna: Wait — unsolved? After all the episodes we've done on Lego bricks and raw eggs, a sponge is worse? Lucas: In some ways, yes. Let me explain. A sponge is soft, porous, compressible, and when wet it changes mass and surface properties. Your typical vacuum gripper or parallel-jaw gripper — the workhorses of industrial robotics — just don't work. A vacuum gripper tries to suck air through the sponge. A parallel jaw crushes it, but can't maintain a stable grip because the sponge squishes out. And to make matters worse, the sponge's shape changes as you apply force, so the robot can't rely on a known geometry. Luna: I know you love a good number. What kind of force are we talking about to deform a sponge? Lucas: Roughly five to ten Newtons — that's about one to two pounds of force — fully compresses a typical cellulose sponge. But to grip it without crushing it, you need to apply less than one Newton. That's the precision challenge. Most industrial force-torque sensors are accurate to about 0.1 Newtons, which sounds fine, but the sponge's deformation is highly nonlinear. The moment you start squeezing, the force jumps fast. So the control loop has to react in milliseconds. And most robot arms are not designed for that kind of delicate, variable-force interaction. Luna: So are there any robots that can handle a sponge? Lucas: Yes, in research labs. The German Aerospace Center — DLR — has a tactile gripper called the dlr hit Hand II that uses arrays of capacitive tactile sensors. Each finger has dozens of taxels — tactile pixels — that measure pressure distribution. They've demonstrated gripping a wet sponge, a raw egg, even a lightbulb. But here's the catch: that gripper costs about fifty thousand euros, and the control software is custom. It's not robust enough for a factory floor where you have to pick up millions of sponges a year. Luna: Right. And I think that points to a bigger issue — the gap between lab demos and industrial deployment. Lucas: Exactly. And that gap is the whole reason Amazon, for example, still uses human pickers for certain items. At Amazon's robotic fulfillment centers, they have hundreds of thousands of robot arms from companies like Fanuc and Kuka. But those arms handle bins, totes, and boxes. When it comes to soft goods — pillows, towels, sponges — humans are still faster and more reliable. Amazon's 2023 research on 'bin picking' showed that for deformable objects, success rates hover around 60 to 70 percent. That's not good enough for a warehouse doing millions of picks a day. Luna: And that 60 to 70 percent — is that for a sponge? Or just any soft object? Lucas: That was for a range of 'soft and deformable' items — sponges were among the worst. The fundamental issue is that a sponge's material properties aren't constant. When it's dry, it's relatively stiff. When it's wet, it's compliant and heavy. The robot has no way to sense that in advance. So you'd need either a vision system that can estimate material properties — which is still early stage — or a gripper that can adapt on the fly. Luna: Speaking of adapting on the fly — what about underactuated grippers? The ones with flexible fingers and cables? Lucas: Good point. Underactuated hands, like the Yale OpenHand project or Robotiq's adaptive gripper, can conform to irregular shapes. They use fewer motors than degrees of freedom, so the fingers passively wrap around an object. And they work great for apples or boxes. But with a sponge, the fingers just push into the sponge without getting a good purchase. The sponge is too soft to trigger the passive mechanics. So the gripper either doesn't close enough, or it closes fully but doesn't exert enough lateral force to hold it. Luna: So it's almost like the sponge is too compliant for the gripper to 'feel' it. Lucas: Exactly. And that's the core research challenge: closed-loop force control with tactile feedback. A robot needs to feel the sponge, adjust grip force in real time, and maintain that force as the sponge deforms or even leaks water. There's a paper from MIT's CSAIL in 2024 that used a gel-based tactile sensor called GelSight to measure shear forces during grasping. They got a sponge — just one specific sponge — to a 90 percent success rate. But again, lab conditions. The sponge was dry, the workspace was controlled. Luna: So the wet sponge is the real villain. What happens when it's wet? Does it change the coefficient of friction? Lucas: Dramatically. Water acts as a lubricant. The coefficient of friction between a typical rubber gripper pad and a dry sponge might be around 0.8. Wet, it drops to about 0.3 or lower. So not only does the sponge become heavier — water can triple its weight — but it also becomes slippery. And because the sponge is porous, water seeps out when squeezed, further reducing friction. It's a moving target. Luna: That sounds like a nightmare for any sort of quality control. Imagine you're packaging cleaning supplies and the robot has to pick up a wet sponge from a conveyor belt. Lucas: It's exactly that application where the problem is most acute. Companies like Procter and Gamble or SC Johnson that make household cleaning products have huge automated lines for bottles and caps, but for sponges they still rely on manual labor. I spoke to a packaging engineer at a trade show last year who said they tried a Fanuc LR Mate with a custom gripper — spent about six months on it — and eventually gave up. They said the sponge's variability was too high. Different cuts, different moisture levels, different storage conditions. Luna: That's a lot of money and time for something that costs a few cents to make. Lucas: Exactly. And that brings us to the economics. A typical industrial robot arm costs about fifty thousand dollars. Integrating a tactile sensor and force control software might add another twenty to thirty thousand. The per-unit cost of picking a sponge becomes prohibitive unless you can reuse that capability for hundreds of other soft objects. But right now, the technology is so specialized that it's hard to justify the investment. Luna: So what's the path forward? Is there any commercial product that's close? Lucas: A few startups are trying. One is Soft Robotics Inc. out of Massachusetts — they make a gripper called the mGrip that uses pneumatically actuated fingers. It's designed for food handling, like picking up chicken tenders or pastries. Those are also soft and irregular. I've seen videos of it picking up a wet sponge, but the success rate isn't published. Another approach is using electrostatic adhesion — basically a flexible pad that creates an electric field to stick to objects. That can work on dry sponges, but moisture kills the effect. Luna: Quick honest thing — a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these deep dives. If today's tech conversation gave you something usable, that's the simplest way to keep it ad-free. Lucas: Yeah, we don't run sponsors, so listener support is what keeps us going. Means we can spend time talking about obscure but fascinating problems like this one. Luna: And we love it. Okay, back to sponges — you mentioned electrostatic adhesion. Any other wild ideas? Lucas: One research direction that's gaining traction is using microspines — tiny hooks that catch on porous surfaces. Think of it like Velcro. For a sponge, microspines can actually snag into the pores and provide a mechanical grip without squeezing. Researchers at Stanford's Biomimetics lab demonstrated a microspine gripper that could pick up a wet sponge with a success rate around 85 percent. But the downside is durability: the spines wear down after a few hundred grasps, and they can damage the sponge surface. Luna: So you either damage the product or the gripper. Not great for manufacturing. Lucas: Exactly. And that's the trade-off. The ideal gripper would combine tactile sensing, variable stiffness, and a gentle touch. There's some promising work using jamming — like a balloon filled with coffee grounds that hardens when you apply a vacuum. That can conform to a sponge and then lock in shape. But the cycle time is slow — about a second to jam and unjam — and in a high-speed line, you need three to five picks per second. Luna: So speed is another constraint. It's not just about gripping, but gripping fast. Lucas: Right. And that's often overlooked in academic papers. They demonstrate a single successful grasp in slow motion. In industry, you need reliability at speed. The sponge problem is a microcosm of a larger challenge in robotics: handling the unstructured, unpredictable world. We've mastered the hard and the rigid. But the soft, the wet, the variable — that's the final frontier. Luna: And we're only beginning to see commercial solutions. I think in the next five years we'll see the first production-ready gripper that can handle a wet sponge. Probably from a startup, not a big robot maker. Lucas: I'd agree. The big robot companies — Fanuc, ABB, Kuka — they focus on high-reliability, high-speed applications where the object is well-defined. The soft stuff is a niche, but it's a growing niche. E-commerce and food processing are driving demand. And as sensors get cheaper — like the Intel RealSense depth cameras that now cost under three hundred dollars — we'll see more integration. Luna: So what should our listeners take away from this? Besides never taking a sponge for granted again? Lucas: I think it's this: the hardest problems in robotics aren't about brute strength or precision. They're about perception and adaptation. A sponge is a simple object, but it contains an entire universe of complexity — variable stiffness, porosity, friction, weight. Until robots can handle that, there's still a massive gap between what they can do and what humans can do. And that gap is where the next breakthroughs will happen. Luna: And maybe the next time you wash dishes, think about the robot that can't. Lucas: Right. Thanks for listening, everyone.