Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Cant Handle a Single Lego Brick
Transcript
- Lucas: So you know those little plastic bricks that have built an empire out of click-together play? Lego bricks. Luna: I have stepped on enough of them in the dark to know them intimately, yes. Lucas: Right. Well, it turns out that picking up one single Lego brick, placing it precisely onto another, and having it actually click — that is genuinely hard for a robot. Harder than you'd think. Luna: Harder than folding a towel? We've done a whole series on impossible robot tasks. Lucas: Different kind of hard. Towels are deformable — they flop. Lego is rigid, but the tolerances are brutal. The clutch power — the force that holds two bricks together — is designed to be about 0.2 millimeters of interference. That's the gap between the stud on top and the tube inside. Luna: Wait — so the brick literally has to deform slightly to lock in? Lucas: Exactly. Lego bricks are made of ABS plastic, which gives just enough. But a robot arm, even with a force-torque sensor, has to apply exactly the right pressure at exactly the right angle. Too hard and the brick skids. Too soft and it doesn't click. Off by a millimeter and the brick sits crooked. Luna: And that matters because... Lego actually makes a lot of its own bricks in factories. Are they using robots? Lucas: They are, but not for the final assembly. Lego factories in Billund and Nyíregyháza use highly customized automation for sorting and packaging. But the actual clicking together? That's still mostly done by humans in some steps, or by purpose-built machines that are essentially jigs — not general-purpose robot arms. Luna: So why is it so hard for a general-purpose arm? I mean, we've seen robots assemble car engines. Lucas: Car engine parts are large, rigid, and tolerances are measured in tenths of a millimeter — but the parts are also gripped firmly. A Lego brick is small, smooth, and you can't just squeeze it hard because you'll damage the clutch. Plus, the alignment needs to be within about 0.1 millimeters for the click to feel right. Luna: And most industrial robot arms have repeatability of about 0.02 to 0.05 millimeters. That should be enough, right? Lucas: In theory, yes. But repeatability is about the robot returning to a taught position. Picking a randomly oriented brick from a bin — that's perception and control. The robot has to see the brick, estimate its pose, plan a grasp, and then execute with compliance. Any error in the vision system, any calibration drift, and the brick doesn't seat properly. Luna: So the challenge is really about sensing and control loops, not just brute force. Lucas: Exactly. There's a famous paper from the MIT CSAIL lab a few years ago where researchers tried to get a robot to assemble a Lego structure from a pile of bricks. They used a force-sensitive gripper and a vision system. The success rate for placing a single brick was around 60 percent on the first try. Luna: That's terrible. A human child can do it blindfolded. Lucas: Right. And the interesting thing is that the failures weren't because the robot couldn't pick up the brick — it could. The failures were in the alignment phase: the brick would be slightly tilted, or the force feedback would misread the contact, and the robot would push the brick sideways instead of down. Luna: So the real problem is that the robot doesn't have the tactile sense to feel the 'click' the way we do. Lucas: That's a big part of it. Humans have thousands of mechanoreceptors in our fingertips. We can feel the brick start to seat, and we adjust micro-movements in milliseconds. A robot's force-torque sensor is usually at the wrist, not the fingertips — so it's sensing the force at a distance, through the gripper, through the brick. The signal is noisy. Luna: So what's the state of the art now? Are there robots that can reliably place a Lego brick? Lucas: There are research systems that achieve over 90 percent success, but they use specialized grippers with embedded tactile sensors — sometimes GelSight sensors, which are like a camera with a soft gel pad that maps pressure. And they use reinforcement learning to train the placement policy. But those systems are not production-ready. Luna: And Lego itself — do they care about this? I mean, they're not exactly hurting for sales. Lucas: They actually do care, because labor costs are rising, and they want more flexible automation. Lego has a robotics lab in Billund that explores using collaborative robots for tasks like sorting bricks by color, or picking bricks from a moving conveyor. But for the final assembly of, say, a Lego set — putting the minifigure in the car, clicking the roof on — they still rely on human hands in some cases. Luna: So the humble Lego brick is a benchmark for robotic dexterity. That's kind of wonderful. Lucas: It is. And it's not just academic. If a robot can reliably pick and place a Lego brick, it can probably handle a lot of small-parts assembly in electronics, medical devices, and consumer goods. Lego bricks are a stand-in for the broader challenge of precision assembly. Luna: And the manufacturing industry would love to automate those tasks. But we're not there yet. Lucas: Not for general-purpose arms. However, there's a company called RightHand Robotics that does pick and place for e-commerce — they can handle thousands of different items, but they still struggle with very small, interlocking parts. Lego is the edge case. Luna: Okay, so if I want to see a robot fail in real time, I just need to give it a Lego set and watch it try to build the Death Star. Lucas: Pretty much. And that's actually a fun thought experiment — because if you watch a robot build a Lego set on YouTube, you'll see it's painstakingly slow. Each brick placement takes many seconds. A human can do it in a fraction of a second. Luna: But the robot doesn't get bored. It could do it for 24 hours straight. Lucas: True. And that's the trade-off. Speed versus endurance. For now, the human still wins on dexterity and speed for novel tasks. But for repetitive, high-volume assembly, purpose-built machines are faster. Luna: I want to go back to something you said earlier — about the clutch power tolerance being 0.2 millimeters. That seems incredibly tight for a toy. Lucas: It is. Lego is obsessive about quality. They mold their bricks with a tolerance of about 0.005 millimeters on the studs. That's why a brick from 1978 still fits a brick from 2026. The clutch power is engineered to be consistent worldwide. And that consistency makes it a perfect torture test for robots. Luna: So when a robot fails to place a brick, it's often because the brick itself varies slightly within that tolerance, and the robot can't adapt. Lucas: Exactly. The variation is small — thousandths of a millimeter — but it's enough to throw off a blind force-controlled placement. Humans can feel the subtle difference and adjust. Robots, especially with wrist-mounted sensors, have a harder time. Luna: Is there any research that tries to give robots a sense of 'click'? Lucas: Yes. Researchers at the University of California, Berkeley, led by Pieter Abbeel, have worked on using audio feedback — the sound of the click — to guide placement. They trained a model to associate the acoustic signature of a successful click with the force profile. That improved success rates significantly. Luna: So the robot listens for the click, like a human would? Lucas: Roughly. They used a contact microphone attached to the gripper. And it worked — the robot could tell, within milliseconds, whether the brick had seated properly, and if not, it would reattempt. That's a clever workaround. Luna: Because vision alone can't confirm the click. You can't see if the connection is solid. Lucas: Right. And that's the deeper insight — manipulation isn't just about position and force. It's about multimodal sensing: touch, sound, even vibration. Humans use all of them. Robots are only starting to integrate multiple senses. Luna: And that's really the frontier of robotic manipulation right now. Lucas: It is. And if you're interested in how these technologies develop, episodes like today's are exactly why we keep this podcast ad-free — we want to explore these nuances without interruption. If you find value in these deep dives, you can support the show at buy me a coffee dot com slash fexingo. It's a small gesture that helps us keep the focus on the robotics and not on sponsors. Luna: Yeah, and it really does make a difference. We hear from listeners who appreciate the depth, and that's what keeps us going. Lucas: So back to the Lego brick: one promising direction is using soft grippers. A soft gripper can conform to the brick shape and apply distributed force, reducing the risk of skidding. That could be a game-changer. Luna: And it mimics the human finger's ability to conform. Lucas: Exactly. Researchers at Harvard's Wyss Institute have developed soft grippers that use pneumatic actuation to wrap around objects. They can pick up a raw egg without breaking it, or a Lego brick with enough compliance to seat it properly. Early results show success rates above 80 percent. Luna: But soft grippers are slower, right? They need to inflate and deflate. Lucas: They are slower. But for precision tasks, speed is secondary to reliability. In a factory, you'd rather have a 99 percent success rate at a slower pace than a 60 percent rate at high speed — because failures cause jams and downtime. Luna: So the future of Lego assembly might be soft grippers with audio feedback and tactile sensing. Lucas: That's the research direction. And it's not just Lego — it's any small-part assembly. Think of watchmaking, electronics, surgical robotics. The Lego brick is the perfect test bench because it's cheap, standardized, and demanding. Luna: I love that a children's toy is pushing the boundaries of robotics. Lucas: It's a beautiful example of how constraints drive innovation. Lego's tight tolerances force us to build better sensors, better control algorithms, better grippers. And every breakthrough on the Lego bench translates to real-world manufacturing. Luna: So next time I step on a Lego brick in the dark, I'll think of it as a tribute to robotic dexterity research. Lucas: And maybe a reminder that we still have a long way to go before robots can do what a child does without thinking. Luna: Alright, that's it for today. Thanks for listening.