Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Pick Up a Slippery Fish
Transcript
- Lucas: So picture this: you're at a fish market, you buy a whole mackerel, the vendor hands it to you — and you grab it. No problem, right? Your fingers automatically adjust pressure, account for the slime, the wetness, the fact that the fish is kind of floppy. Luna: And if it starts slipping, you instinctively grip harder or shift your hand. Lucas: Exactly. Now ask a robot arm to do that. And I mean any robot arm — the best industrial gripper you can buy, a collaborative robot from Universal Robots, even a research-grade robot hand like the Shadow Hand. They all fail. A wet, slippery, deformable fish is basically kryptonite for current robotics. Luna: We've done a lot of episodes on things robot arms can't handle — coins, fitted sheets, wet soap. But fish feels like it's in a class of its own. Lucas: It really is. And the reason goes beyond just 'slippery.' It's a combination of three factors: low friction when wet, non-rigid body that deforms under pressure, and the fact that the surface is covered in a layer of mucus and water that makes traditional grippers lose purchase. This is episode 119, and by the way, if this conversation is giving you something new to think about, that's exactly why we keep the show ad-free. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Luna: Yeah, it's a small gesture that helps us keep digging into these niche problems that most people don't even know exist. Lucas: Right. So back to the fish. I looked up the numbers: the global seafood processing industry handles something like 180 million tons of fish per year. A huge portion of that is still done by hand because automation can't reliably pick up individual fish without damaging them. Luna: And damage matters. A bruised fish is worth less. So they're not just trying to grab the fish — they're trying to grab it gently. Lucas: Exactly. And that's where the research is at. One of the most promising approaches comes from MIT's Computer Science and Artificial Intelligence Lab — they've developed a 'visuotactile' sensor called GelSight. It's essentially a soft, gel-like pad with a camera embedded inside. When the gripper touches an object, the gel deforms, and the camera captures the deformation in high resolution. The robot can then 'see' the contact area and adjust its grip in real time. Luna: So it's like giving the robot a sense of touch through vision? Lucas: Exactly that. And it works surprisingly well on slippery objects. In a 2025 paper, the MIT team demonstrated a gripper using GelSight that could pick up a wet fish — specifically a raw salmon fillet — without dropping it, and without damaging it. They achieved a 96 percent success rate over 200 trials. Luna: That's impressive, but salmon fillet is a bit more structured than a whole mackerel. Have they tried whole fish? Lucas: They have. The success rate dropped to about 82 percent for whole fish. The problem is that whole fish have irregular shapes — fins, tail, head — and the gripper has to find a stable grasp point. But 82 percent is still way better than any previous system. And it's improving fast. Luna: What about other approaches? I know some companies are using suction cups. Lucas: Suction cups work well on smooth, non-porous surfaces. But fish are wet, and the slime creates micro-leaks. Plus, if the surface is uneven, suction fails. There's also the 'freeze-gripper' approach — basically a cold surface that freezes a thin layer of water between the gripper and the fish, creating an ice bond. It works, but it's slow and consumes energy. Luna: And you can't freeze a fish that's going to be sold fresh. Lucas: Right, that destroys the texture. So visuotactile sensing seems like the most practical path. But there's another really interesting line of research coming out of the University of Tokyo — they're using 'electroadhesion.' You apply a voltage to the gripper, which creates an electrostatic attraction between the gripper and the object. It works on wet surfaces because water is polar. Luna: Wait, so you're essentially gluing the fish to the gripper with static electricity? Lucas: More or less. The force is relatively weak — about the same as a light pinch — but for a one-kilogram fish, it's enough if you have enough surface area. The challenge is that it requires a very flat, clean contact. If there's a lot of slime or scales, the adhesion drops. So right now, it's more of a lab curiosity than a commercial solution. Luna: But it's fascinating that there are so many different ways to attack the problem. Lucas: And the commercial pressure is real. I spoke to a sourcing manager at one of the big Norwegian salmon processors — they're desperate for automation. Labor is scarce, especially in remote processing plants. They're willing to pay a premium for any system that can handle fish reliably. That's driving a lot of R&D. Luna: So what's the timeline? When might we see a robot arm that can reliably pick up a fish in a commercial setting? Lucas: I'd say three to five years for a dedicated fish-handling robot. The GelSight technology is being commercialized by a spin-off called GelSight Inc. They already have industrial sensors for quality inspection. Adding grasping is the logical next step. But for a general-purpose robot that can pick up anything — fish, a wet bar of soap, a slippery piece of fruit — that's probably a decade away, if not more. The reason is that each object requires a different strategy. Fish are slippery but somewhat forgiving because they're soft. A wet wine glass is a completely different challenge — rigid, fragile, slippery. Luna: And we haven't even talked about live fish, which would add the problem of motion. Lucas: Live fish are a whole other nightmare. But interestingly, the seafood industry doesn't need that. Most processing happens after the fish is stunned or killed. So the motion problem is less urgent. Luna: Still, the fact that a simple act like picking up a fish is so hard for robots really puts into perspective how amazing human hands are. Lucas: Absolutely. Your hand has about 27 degrees of freedom, hundreds of mechanoreceptors per square centimeter of skin, and a real-time control system that's been fine-tuned by evolution for millions of years. A robot gripper might have two degrees of freedom and one pressure sensor. It's a fundamentally different capability. Luna: So the next time someone complains that their robot vacuum still bumps into furniture, you can remind them: at least it's not trying to pick up a slippery fish. Lucas: Exactly. That's the bar. And honestly, it's a high one. But we're making progress. The MIT work, the electroadhesion work — these are real steps toward a future where robots can handle the messiness of the physical world. And when they can, it'll transform industries from seafood to surgery. Luna: Surgery? Are you thinking of surgical robots handling slippery tissue? Lucas: Exactly. The same visuotactile sensing could help a surgical robot grasp a slippery organ or blood vessel without damaging it. In fact, there's already a research group at Johns Hopkins adapting GelSight for minimally invasive surgery. So the fish problem isn't just about fish — it's about a whole class of objects that are wet, deformable, and fragile. Solve that, and you unlock a huge range of applications. Luna: That's a great way to frame it. And it makes me wonder: what's the next 'impossible' grasp after fish? Lucas: I think it's something like a raw egg — fragile, slippery, and deformable. Some labs have done it with custom grippers, but not reliably. Or maybe a live crab — hard shell but with legs that move unpredictably. That would combine grasping and stability in a really interesting way. Luna: Sounds like a future episode. Lucas: We'll add it to the list. For now, the takeaway is: if you want to appreciate the genius of your own hand, try to build a robot that can pick up a fish. It's harder than it looks.