Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Peel a Potato
Transcript
- Lucas: So, today we are talking about something that sounds incredibly mundane but is actually a nightmare for most robot arms: peeling a potato. Luna: A potato. Really? I mean, we've talked about eggs, feathers, deflated balloons. How is a potato harder? Lucas: Well, on the surface, it seems simple — you hold a potato in one hand and a peeler in the other. But for a robot, it's a multi-dimensional problem because the potato is irregular, it's slippery, and the peel thickness varies. And unlike cutting or gripping, peeling requires sliding contact with variable force. Luna: Right, so it's not just about grabbing it. It's about maintaining the right pressure while moving the tool along a curved surface. Lucas: Exactly. And if you think about it, this is a classic example of what roboticists call 'dexterous manipulation' — tasks that require continuous sensory feedback and adaptive control. And a potato, of all things, has become a benchmark for this. Luna: I had no idea. So how do current robots handle it, if at all? Lucas: Well, let me give you a specific example. There's a company called DexRob — small startup out of Boston — and they've been working on a robotic system specifically for peeling potatoes. They use a soft robotic gripper with embedded tactile sensors, and they've trained a neural network to adjust the peeling angle and force in real time. But even with all that, their system takes about twice as long as a human and still bruises about 15% of the potatoes. Luna: Fifteen percent. So not quite ready for the commercial kitchen. Lucas: Not yet. But the interesting thing is that this research is pushing the boundaries of tactile sensing. MIT's soft robotics lab has been working on a similar problem, and they've developed a sensor that can measure shear force and pressure simultaneously — which is crucial for peeling. And that technology has applications way beyond potatoes: think surgical robots, handling delicate electronics, even fruit harvesting. Luna: It's funny — something as simple as peeling a potato could lead to breakthroughs in so many fields. Lucas: And that's exactly why we're talking about it today. Because the humble potato is actually a pretty tough test for robotic dexterity. And if you find these kinds of deep dives into robotics useful, you know, a couple of dollars a month is genuinely what keeps these going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. Luna: Yeah, it really helps us keep the show ad-free and focused on the details. Lucas: So back to the potato. One of the reasons it's so hard is that the potato's shape is not just irregular — it's also anisotropic. The skin thickness varies depending on where on the potato you are, and there are eyes and blemishes that change the surface properties. Luna: So the robot has to constantly adjust its strategy mid-motion. Lucas: Exactly. And that's where tactile sensing comes in. Most industrial robots operate in a 'position control' mode — they go to a specific coordinate. But for peeling, you need force control and impedance control, so the robot can 'feel' the surface and adapt. Luna: So what does DexRob's system actually look like? I'm imagining a robot arm with a peeler attachment. Lucas: That's basically it. They use a standard six-axis arm from Fanuc — the CRX-10iA — but they've replaced the end effector with a custom soft gripper that has a peeler mounted on a passive compliant joint. And the gripper itself has 16 tactile sensors on each finger, measuring force and slip. Luna: So the robot holds the potato in one gripper and the peeler in the other? Like a human would? Lucas: Actually, no. In their setup, the robot holds the peeler, and the potato is held by a passive fixture. But MIT's approach is different — they have two arms: one holds the potato, the other holds the peeler. That's harder because both arms have to coordinate. Luna: I can see why that would be a challenge. So what's the state of the art? How close are we to a robot that can peel a potato as well as a person? Lucas: Well, in a controlled setting, DexRob's system can peel a potato in about 45 seconds, with about 10% waste. A skilled human can do it in 20 seconds with 5% waste. So we're still not there, but the gap is closing. And the key insight is that the robot's performance is limited by sensing, not by actuation. Luna: Meaning the hardware is good enough, but the software and sensors aren't quite there yet. Lucas: Right. The motors and joints can move precisely, but the robot doesn't 'feel' the potato the way we do. Our skin has millions of mechanoreceptors; a robot might have 16. So it's a data problem as much as a control problem. Luna: So what's the next step? More sensors? Lucas: That's part of it. Researchers are developing high-density tactile skins that can be wrapped around robot fingers, giving them thousands of sensing points. But there's also the software side — better algorithms for interpreting that sensory data and making split-second adjustments. Luna: It sounds like the potato is becoming the new 'grasping a ball' benchmark for dexterity. Lucas: Exactly. And there's a reason for that. Potatoes are cheap, readily available, and they offer a consistent but variable challenge. Plus, if you can solve potato peeling, you can probably solve a lot of other food processing tasks — like peeling apples, carrots, or even fish. Luna: So beyond the kitchen, what industries would benefit from this? Lucas: Think about any task that involves handling fragile, irregular objects. That includes everything from electronics assembly — where you might need to pick up a delicate chip — to medical robotics, where you need to manipulate tissue. The same underlying technology could be used for robotic surgery, like peeling a grape in a simulated procedure. Luna: I've seen videos of robots peeling grapes. It's impressive, but it's usually in a lab setting with a single grape. Lucas: Right, but the principles are the same. And some researchers have actually used grapes as a proxy for surgical tasks. But potatoes are more challenging because they're larger, more irregular, and have a tougher skin that requires more force. Luna: So where does this leave us? Are we going to see robotic potato peelers in commercial kitchens anytime soon? Lucas: I think we're maybe five to ten years away from a commercially viable system. The cost of the sensing and computation is still high, and the reliability isn't there yet. But there's a lot of interest from food processing companies — especially ones that deal with large volumes, like frozen food manufacturers. Luna: So it's not just about replacing a kitchen worker; it's about scaling up operations. Lucas: Exactly. A robot that can peel potatoes 24/7, even if it's a bit slower than a human, could be a huge advantage in a factory setting. And as the technology improves, the cost will come down. Luna: So the potato is really a testbed for a whole class of manipulation problems. Lucas: That's exactly right. And it's a great example of how a seemingly simple task reveals the complexity of the physical world. Our brains and hands are incredible — we don't even think about peeling a potato. But for a robot, it's a grand challenge. Luna: Well, I'll never look at a potato the same way again. Lucas: And maybe that's the takeaway — that the next big breakthrough in robotics might come from solving something as mundane as dinner prep. Luna: Alright, thanks for diving into the spud with us. Lucas: Indeed. Until next time.