Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Hands Can't Handle a Raw Egg
Transcript
- Lucas: We spend a lot of time on this show talking about things robots can't do — and today's might be the most relatable yet. Luna: You've intrigued me. Not another USB plug situation, I hope. Lucas: Worse in some ways. I'm talking about picking up a raw egg. You know, the thing you do every time you make an omelette without even thinking about it. Luna: Okay, I see where this is going. That's a classic challenge in manipulation — fragile, deformable, non-uniform shape. Lucas: Exactly. And the reason I wanted to drill into this today is that there's a really concrete benchmark that came out of MIT's Soft Robotics Lab last year. They tested a current-generation tactile sensor array — essentially a robotic fingertip with hundreds of pressure sensing points — against supermarket eggs. Luna: How'd it do? Lucas: About a 30 percent failure rate across a hundred trials. Either the egg slipped or it cracked. And these were medium-sized, standard grocery store eggs, not some weird shape. Luna: That's surprisingly high. I would have guessed maybe 10 percent, given the advances we've seen in force control over the last five years. Lucas: Right — and that's the thing. Force control has gotten really good at knowing how much pressure is being applied in a static hold. But an egg is dynamic. The shell is a thin, brittle arch structure. The moment you apply any shear force — like when you start to lift — the stress distribution changes instantly. Luna: So the sensor is reading the force at contact points, but the egg's shell is distributing that stress in ways that aren't linear or easily predictable. Lucas: Exactly. And the shell thickness varies by about 0.3 millimeters even on a single egg. So the robot doesn't know if it's gripping a thick part or a thin part. The MIT team found that most cracks initiated not at the contact point, but about 5 millimeters away, along a line of stress concentration. Luna: That's a structural engineering problem as much as a robotics one. It reminds me of why it's so hard to design a robot that can pick up a potato chip without shattering it — same issue, different scale. Lucas: Precisely. And the egg is a great proxy for a whole class of objects: fruit, glassware, ceramics, even human tissue in surgical robotics. If you can reliably handle an egg, you can handle a lot of delicate things. Luna: So what are the approaches to solving this? More sensors? Better algorithms? Lucas: There are two main camps. One says — more sensors, higher resolution, faster feedback loops. So instead of hundreds of pressure points, you have thousands, and you sample at kilohertz rates. The other camp says — forget rigid sensors, go biomimetic. Use soft, compliant materials that deform around the object, like a human hand does. Luna: I've seen some work from the University of Tokyo on a gel-based gripper that molds to the shape of the object. Is that still active? Lucas: Very much so. In fact, there was a paper just last month from a group at Stanford — they used a silicone-based gripper with embedded air channels. When it contacts an egg, it inflates slightly to create a conformal grip, then deflates to release. They reported a 94 percent success rate on raw eggs. Luna: That's impressive. But I imagine the trade-off is speed — pneumatic actuation isn't exactly snappy. Lucas: Right — cycle time is about three seconds per pick and place. In an industrial setting, that's too slow. A human packer can do an egg in under a second. So the challenge becomes: can you get the compliance of soft robotics with the speed of rigid grippers? Luna: Is anyone trying a hybrid? Like a rigid skeleton with a soft, sacrificial outer layer? Lucas: Yes — there's a company called Soft Robotics Inc. that makes a gripper with flexible fingers and a rigid backbone. They've been testing it in poultry processing plants for handling eggs without breaking them. Their literature claims 98 percent reliability, but that's under controlled conditions — consistent egg size, temperature, humidity. Luna: So the real world is still messy. A warm egg from a hen is different from a cold one straight out of the fridge. Lucas: And that's the rub. Most lab tests use room-temperature eggs. But in a commercial kitchen or a farm packing line, you get temperature gradients, condensation on the shell, variations in surface friction. The MIT study I mentioned actually tested both room-temp and refrigerated eggs — the failure rate doubled for cold eggs because condensation made the shell more slippery and the gripper couldn't adjust its friction model fast enough. Luna: That's fascinating — and it ties back to a broader theme we've touched on before: robots struggle with edge cases that humans handle without thinking. Lucas: If today's conversation gave you something useful — maybe a new appreciation for why your breakfast is still prepared by human hands — the way these episodes stay ad-free is through listener support. You can buy me a coffee dot com slash fexingo. That's buy me a coffee dot com slash fexingo. It's a small gesture that keeps the show independent and focused on the stuff that actually matters. Luna: Absolutely. And it's because of listeners like that that we can spend an entire episode on the physics of an egg without having to cut for a commercial break. Lucas: Alright — back to the breaking point. So where is the field headed? I think the next frontier is not just gripping, but sensing the internal state of the object. Some labs are experimenting with acoustic resonance — tapping the shell and listening to the sound to gauge thickness and integrity before applying force. Luna: Like a doctor percussing a patient's chest. That's clever — it uses the object itself as part of the sensor. Lucas: Exactly. A group at Carnegie Mellon has a prototype that does a quick tap — less than 10 milliseconds — and uses a small microphone to analyze the frequency response. They can detect whether an egg is cracked before they even try to pick it up. Luna: That's prehensile diagnosis. It changes the problem from 'grip it right' to 'decide if you should grip at all'. Lucas: Right. And that kind of decision-making is where machine learning comes in. The same group trained a neural network on thousands of tap sounds from eggs with known crack conditions. It can now classify an egg as intact or cracked with 98.5 percent accuracy in under 50 milliseconds. Luna: So the robot can reject a cracked egg before attempting a pick. That avoids the mess and the contamination risk. Lucas: Exactly — and in a food processing line, that's huge. One cracked egg can contaminate an entire batch of liquid egg product. The USDA estimates that costs the industry about 30 million dollars a year in recalls and waste. Luna: Wow — so this isn't just a lab curiosity. There's a real economic incentive to solve it. Lucas: Absolutely. And that's why I think we'll see commercial egg-handling robots within the next three to five years. The technology is converging — better tactile sensing, faster soft actuation, and now acoustic pre-screening. The missing piece is integration and cost reduction. Luna: What about the ethical side? If robots take over egg packing, what happens to the workers who do it now? Lucas: That's a fair question. In many countries, egg packing is already highly automated for grading and sorting. The manual part is usually the actual handling — placing eggs into cartons. That's a repetitive motion task that can cause strain injuries over time. So there's a case to be made that automation could reduce workplace injuries. But it's also a job that pays relatively well in rural areas. The transition needs to be managed. Luna: I suppose it's never just a technology problem. It's always a people problem too. Lucas: Right. And that's why we talk about it here — not just the mechanics, but the context. So next time you crack an egg, maybe spare a thought for the robot that still can't do it. Luna: And maybe appreciate your own opposable thumbs a little more. Lucas: Exactly. That's all for today. Until next time.