Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Hands Still Can't Grip a Slippery Fish
Transcript
- Lucas: So there's a YouTube video that's been making the rounds in robotics labs—it's about fifteen seconds long, and it shows a robotic gripper trying to pick up a raw fish fillet from a conveyor belt. And it fails. It fails for about twenty minutes straight. Luna: Twenty minutes of a robot fumbling a fish. That's oddly mesmerizing to watch, I bet. Lucas: It is, but it's also a perfect illustration of one of the hardest problems in manipulation: gripping wet, slippery, deformable objects. And we're not just talking about fish—think of a bar of soap, a peeled mango, a water bottle covered in condensation. The robot world has cracked rigid objects, but this whole category of 'squishy and slick' is still a frontier. Luna: I've seen robots pack boxes of canned goods and even handle eggs, but a raw chicken fillet or a fish—that just seems like a different universe. What makes it so hard? Lucas: A few things stack up. First, friction. When a surface is wet, the coefficient of friction drops dramatically—like, your fingertip on a wet plate versus a dry one. Second, deformation. A fish fillet isn't rigid; it bends, squishes, and changes shape the moment you touch it. And third, the geometry is inconsistent—no two fillets are the same, so the gripper can't just rely on a pre-programmed shape. Luna: So it's not just about adding more force or better sensors. It's a materials problem at the contact point. Lucas: Exactly. And that's what a few labs and startups have been focusing on. One approach is to mimic nature—specifically, the microstructures on gecko feet. Those tiny hairs create van der Waals forces, which work even on wet surfaces. Researchers have been embedding synthetic versions of those geometries into gripper pads. Luna: So we're literally borrowing from geckos to grip a fish. Lucas: In a way, yes. But the challenge is scaling that up to something that can handle the weight and the slippery protein film on a fish. There's a company in Norway that's been testing these gecko-inspired pads on cod fillets in a processing plant, and they're getting closer, but it's not quite there yet for high-speed production. They're at about ninety percent success rate, which sounds good until you realize that a one in ten drop on a fast line is a costly failure. Luna: And that's a real problem in seafood processing, where robots are desperately needed because of labor shortages. I read that some plants in the US and Europe can't find enough workers to do this kind of repetitive, cold, wet work. Lucas: Right, and that's the economic driver. The global seafood processing market is massive, and automation could save millions, but only if the robots can actually handle the product without smashing it or dropping it. There's also a major safety component—human hands get cut, get cold, get repetitive strain injuries. So it's not just about efficiency; it's about replacing jobs that are genuinely hard for people to do. Luna: But even if you solve the friction and deformation, how does the robot know how hard to squeeze? I mean, a fish fillet can go from firm to mush in a split second if you press too hard. Lucas: That's where the sensing comes in, but not in the way you might think. Standard force-torque sensors measure the total force, but they don't tell you the distribution across the contact. So a new wave of tactile sensors—thin, flexible skins with tiny pressure cells—can map the pressure in real time. A team at MIT has a sensor that's about the size of a postage stamp, and it can detect the exact spot where a slippery object is starting to slip. Luna: So the robot can react before it drops, kind of like when you automatically tighten your grip on a wet glass. Lucas: Exactly. But the tricky part is that the algorithm has to process that tactile data fast enough. In a human, it's a reflex loop—you don't think about it. In a robot, you have to run a model that predicts slip based on micro-vibrations. That's where machine learning comes in, and it's actually showing some promise. In a recent paper, they trained a neural network on thousands of slip events with different wetness levels, and it learned to predict a slip about two hundred milliseconds before it happens. Luna: Two hundred milliseconds—that's not much time, but for a robot, that's a lot of compute. Is that enough? Lucas: In lab tests, yes. The system was able to increase the grip force just enough to prevent the slip, without crushing the object. They tested it on a wet bar of soap and a piece of raw chicken—two of the most hated objects in robotics. And they got a significant improvement in success rate, from about sixty percent up to ninety-four percent. Luna: That's a big jump. But is this just lab magic, or can it work in a real processing plant with all the noise and mess? Lucas: That's the million-dollar question. In a lab, the environment is controlled. In a fish plant, there's water spray, temperature fluctuations, and the fillets are coated in a slippery protein film that just won't wipe off. Some engineers think the real breakthrough will come from combining gecko-inspired surfaces with active suction. There's a startup in Japan that's doing that—they have a gripper with tiny suction cups and microtextured pads, and they've run it on a pilot line for squid, which is even more slippery and deformable than fish. Luna: Squid—wow. That's like the final boss of slippery objects. Lucas: It kind of is. And they're getting some traction. But the thing is, even if you have the perfect gripper, you still have to deal with the fact that every fish is different. So the robot has to adapt on the fly, and that's a control problem, not just a hardware problem. It's a full-stack challenge. Luna: So where does that leave us? Are we years away from a robot that can fillet a fish autonomously? Lucas: I would say we're closer than we were five years ago, but there isn't a single system that can do the whole job. The gripper is the bottleneck, but once you solve that, the rest—the cutting, the trimming—is more like classical automation. The pieces are coming together, but it's a slow grind. Luna: And that's what makes this field so interesting—it's not about one miracle breakthrough, it's about these incremental gains in materials, sensing, and machine learning. Lucas: Exactly. And it's worth noting that the same technology that helps grip a fish fillet also helps in other industries—think handling soft fruit in agriculture, or grasping wet components in a food processing line. The research is flying under the radar, but it has huge commercial potential. Luna: You know, this reminds me—we only get to make deep dives like this because of the listeners who chip in a few bucks a month on buy me a coffee dot com slash fexingo. It's not a big ask, but it literally covers the cost of the studio time and the research we do for each episode. We don't run ads, so that's what keeps this show going. Lucas: Yeah, and it means the world to us. If you find yourself quoting a fun fact about slippery fish to a friend, maybe consider throwing in a coffee. It's a small way to keep this kind of nerdy deep dive alive. Luna: Alright, back to the fish. So we've got the materials and the sensing, but I'm curious—what about the actual grip strategy? Are they trying to mimic human fingers, or is it something completely different? Lucas: Great question. Some teams are going with a more human-like approach—two fingers that apply pressure from opposite sides. But others are trying something entirely different: a sort of 'hand' that wraps around the object like a web. Think of a soft, inflatable bladder that conforms to the shape. That's what the company Soft Robotics has been doing, and they've had success with items like donuts and even chicken tenders. Luna: Oh, I've seen those—they look like a balloon with fingers. They're less precise, but they're great for irregular shapes. Lucas: Right. And the trade-off is that you lose some precision for the ability to handle a wider range of objects. For a fish fillet, you might want a hybrid approach—a firm but compliant grip that can also sense and adapt. There's a group at Harvard that's working on a soft gripper with embedded microchannels for fluidic actuation, and they've been testing it on raw meat. They're getting impressive results, but it's still slow. Luna: Slow makes sense for now. It's like the early days of robot arms in automotive manufacturing—they were slow and expensive, but eventually they became ubiquitous. Lucas: Exactly. And the same will happen here. The market is there, the need is real, and the technology is improving. So the next time you see a robot fumbling a fish on YouTube, just remember—that's not a failure, that's a progress update. Luna: I love that framing. A progress update. So what should we watch for in the next year or two—any specific milestones? Lucas: I'd watch for a commercial system that can handle a full fish processing line—head off, fillet, skin removal—without human intervention. There are a few companies claiming they're close, but I'd like to see a third-party validation. Also, keep an eye on the tactile sensing space; I think that's where the biggest leaps will come. Luna: And if that happens, what's the broader impact beyond seafood? Lucas: It's huge. It opens up automation for a whole range of 'unstructured' tasks—like picking mushrooms, handling fresh herbs, even sorting recycling where objects are wet and variable. It's the difference between robots that work in carefully controlled environments and robots that can work in the messy, unpredictable world. That's the real goal. Luna: So maybe we'll see robots packing our lunch salads someday. Lucas: Someday, and that day might not be as far off as we think. But for now, the fish are winning. And that's okay—it's a good reminder of how hard it is to replicate the human hand. Luna: Well, I'm rooting for the geckos. They seem to have the edge. Lucas: They do. And maybe we'll see a fish-gripping robot that's inspired by them in a fridge near you. That's the kind of future we're looking forward to.