Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Pick Up a Single Grain of Rice
Transcript
- Lucas: So let me start with something that sounds absurd: a robot arm that can assemble an iPhone motherboard can't reliably pick up a single grain of rice. Luna: That does sound absurd. But I have a feeling you're about to tell me it's a surprisingly hard problem. Lucas: It is. And it's not just rice — it's any small, lightweight, irregularly shaped object. Think of a sesame seed, a grain of salt, a tiny screw, a pharmaceutical pill. The same physics that let you pinch a grain of rice between your thumb and index finger are incredibly hard to replicate in hardware and software. Luna: And, you know, if today's deep dive into the rice grain problem gives you something you can actually use — or even just makes you appreciate how amazing your own fingers are — I'll just say: if it was worth a coffee to you, that's the link. Buy me a coffee dot com slash fexingo. Lucas: Yeah, listener support is what keeps this show ad-free and lets us spend time on weird corners like this. So seriously, if today's topic clicks, that's the place. Luna: Alright, back to the rice. So what makes a single grain so hard for a robot? Lucas: Three main challenges. First, size and weight. A grain of rice is about five to seven millimeters long, weighs around 20 milligrams. That's tiny. Most standard grippers are designed for objects that weigh at least a few grams. Second, shape — it's irregular, curved, and not symmetric. Third, surface texture — rice has a slightly sticky outer layer when dry, but it's also brittle. Squeeze too hard and you crush it; too light and you drop it. Luna: So it's the Goldilocks problem applied to micro-manipulation. Lucas: Exactly. And that's where adhesion forces — van der Waals forces, capillary forces — become dominant over gravity at this scale. When an object is that light, electrostatic and moisture-based adhesion can make it stick to the gripper unpredictably. So even if you get a good pinch, the rice might cling to one finger when you try to release it. Luna: Right, because the forces that matter change as objects get smaller. For a car door, gravity is king. For a grain of rice, surface forces are king. Lucas: Precisely. And most robot grippers are designed for the macro world — they use friction-based pinching and assume gravity dominates. They don't account for the fact that at the micro scale, objects can stick to the gripper due to humidity or static charge. Luna: What about tactile sensors? Can't they help the robot 'feel' whether it's holding the rice properly? Lucas: They can, but tactile sensing for sub-millimeter features is still in the lab. The most common tactile sensors — like the GelSight sensor developed at MIT — give high-resolution surface images, but they're bulky. You can't embed them into a fingertip small enough to pinch a single grain. There's work on optical micro-force sensors, but none are commercially mature. Luna: So no off-the-shelf solution exists for this specific task? Lucas: Not really. But there is a recent paper from MIT's CSAIL — the GRAB Lab, led by Professor Alberto Rodriguez — that tackled exactly this. They published it last year, around July 2025. They designed a custom gripper that uses two miniature ceramic-tipped tweezers with a built-in force sensor that can measure forces down to 0.1 millinewtons. Luna: 0.1 millinewtons — that's the weight of about 10 grains of rice? Let me check... a grain of rice exerts about 0.2 millinewtons of force due to gravity. So they can sense forces half that. Impressive. Lucas: Right. And they combined that with a control policy that actively regulates grip force based on real-time tactile feedback. They tested it on 500 individual grains of jasmine rice, and they achieved a 73 percent success rate at picking and placing without damage. Luna: Seventy-three percent — that's good for a lab prototype, but not production-ready. What were the failure modes? Lucas: The main failures were grains that shattered because the force feedback lagged slightly, and grains that stuck to one tweezer tip due to static. They also had a few cases where the rice slipped because the surface moisture varied between grains. So it's not just about the hardware — the control loop needs to adapt to individual grain properties. Luna: Which is what human fingers do unconsciously. We adjust our grip on a grain of rice based on how it feels — is it dry, is it sticky, is it slightly curved? Lucas: Exactly. And that's the frontier of robotic dexterity: not just grasping, but sensing and adapting in real-time. Now, you might ask: why does this matter beyond a neat lab trick? There are real industries that need this. Food processing — think of sorting grains, seeds, or spices. Pharmaceutical — picking and placing individual pills or capsules into blister packs. Electronics assembly — handling tiny capacitors, resistors, or microchips. Luna: In electronics, don't they already use pick and place machines for surface-mount components? Lucas: They do, but those machines use vacuum nozzles, not grippers. Vacuum pick and place works well for flat, smooth components like chips. But for curved or irregular objects like a grain of rice or a seed, vacuum fails because the seal is poor. A gripper with adaptive force control is more versatile. Luna: So the rice grain is a benchmark problem — like a 'Hello World' for micro-manipulation. Lucas: Exactly. And there's a broader trend here. In the last two years, several labs have started publishing on 'granular manipulation' — handling objects that are smaller than a centimeter and deformable. It's becoming a subfield. There was also work from Stanford's CHARM Lab on using electrostatic adhesion to pick up single poppy seeds, and from ETH Zurich on using ultrasonic vibration to reduce adhesion during release. Luna: That's clever — use vibration to shake off the sticking forces. But does that work for brittle objects like rice? Vibration might crack it. Lucas: It depends on the frequency and amplitude. At ultrasonic frequencies, the displacement is tiny — micrometers — so it might not damage the grain. But it's another variable to tune. The point is, there's no universal solution yet. Each object class requires a tailored approach. Luna: Which brings us back to the fundamental question: how do we build a gripper that can handle arbitrary small objects without programming each one? Lucas: That's the holy grail. And some researchers think the answer is not better rigid grippers, but soft robotics — using compliant materials that conform to the object's shape and distribute force evenly. A soft silicone finger can wrap around a grain of rice without crushing it, and if you embed magnetic particles, you can control its stiffness on the fly. Luna: I've seen those — variable stiffness grippers from Harvard's Wyss Institute. They can go from floppy like a jellyfish to rigid like a claw in milliseconds. Lucas: Right. That approach could be ideal for rice-like objects. But soft grippers have their own issues: slower cycle times, lower precision, and difficulty with sensing. So there's a trade-off. Luna: Do you think we'll see a commercial robot that can handle a single grain of rice reliably within, say, five years? Lucas: I think we'll see specialized systems in controlled environments — like a pharmaceutical pick and place cell — within three years. But for general-purpose, open-world rice handling? Probably closer to ten. The sensory and control challenges are deep. Luna: So for now, the humble grain of rice remains a stubborn reminder of how far robotics still has to go. Lucas: And also a reminder of how remarkable the human hand is. Every time you pick up a single grain of rice — whether to eat it or just to move it — you're solving a control problem that the best labs in the world can only crack 73 percent of the time. Luna: That's a humbling thought. And a great point to end on. Thanks, Lucas. Lucas: Thanks, Luna. And thanks to all of you for listening. If this episode made you think differently about your own fingertips, you know where to find us — buy me a coffee dot com slash fexingo.