Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Can't Use a Pair of Scissors
Transcript
- Lucas: So last week I spent three days trying to get a standard six-axis robot arm to cut a piece of printer paper with a pair of scissors. By Friday, the paper looked like a preschool art project. Luna: Oh no. That bad? Lucas: Worse. The robot either squeezed the scissors too tight, locking the blades, or held them so loosely that the blades just pushed the paper around. And when it did manage a cut, the trajectory was off by maybe two millimeters, so the scissors just folded the paper instead. Luna: Right, because the scissors need to both rotate and translate simultaneously — the pivot point moves as you cut. Lucas: Exactly. And that pivot motion is mechanically simple for a human — your thumb and fingers coordinate instinctively — but for a robot, it's a multi-variable control problem. You've got two fingers applying force at different points along the handles, and the axis of rotation moves dynamically. Luna: Plus the material feedback changes. Cardboard, fabric, plastic — each one requires a different force profile. Lucas: Right. A robot can cut a straight line with a laser or a waterjet. But a pair of scissors? That's a closed-loop task that combines vision, tactile sensing, and real-time force control. It's why you don't see scissor-wielding robots on any factory floor. Luna: I'm thinking of surgical robots — the da Vinci system uses scissor-like instruments for cutting tissue. That's got to be easier because the instrument is designed for the robot, right? Lucas: Partly. The da Vinci's instruments have a cable-driven wrist that mimics human motion, but those scissors are purpose-built — they're not off-the-shelf Fiskars. And even there, the surgeon directly controls each movement. Full autonomy with scissors is still research-stage. Luna: So what makes a cheap pair of office scissors so uniquely hard? Let's break down the physics. Lucas: Okay. First, the grasp. You need to hold the handles with two opposing forces — your thumb pushes one handle, your four fingers pull the other. The robot's gripper has to exert enough friction to prevent slipping, but not so much that it deforms the plastic handles. Luna: And the handles are often curved and tapered, so the contact geometry changes as you open and close. Lucas: Right. Second, the pivot joint itself. On most household scissors, the screw that holds the blades together is designed to be loose enough for smooth operation but tight enough to keep alignment. A robot trying to apply force through that joint induces parasitic torques that can jam it. Luna: So when you squeezed too tight, you essentially locked the pivot. Lucas: Exactly. The third thing is the shear point. To cut, the blades must meet at a precise angle along the entire length. If the robot's wrist is off by even a degree, the blades don't align and you get a crease, not a cut. Luna: And that misalignment is dynamic — as you cut, the stiffness of the paper changes the force distribution. Lucas: That's the fourth thing: material interaction. Paper is thin and compliant. The moment the blades bite, the paper buckles unless you hold it taut. But holding taut with a second gripper while the first manipulates the scissors? That requires coordination between two separate arms. Luna: And most industrial setups have only one arm. So you're asking one robot to do a two-handed job. Lucas: Right. And that brings us back to the fundamental gap: fine manipulation with passive tools designed for human hands. It's not just about better grippers or better algorithms — it's about rethinking the tool. Luna: Before we dive into solutions, I want to step back. You know, it's episodes like this that remind me why listener support matters. This show has no ads, no sponsors — it's funded entirely by a small group of listeners who chip in via buy me a coffee dot com slash fexingo. Lucas: Yeah, and honestly, that's what lets us spend three days maiming paper with robot arms and then explain why it's interesting. If you get something out of these deep dives, consider joining that group. It's a tiny gesture that keeps us independent. Luna: Absolutely. Now, back to those scissors — there are actually labs working on robotic hands specifically designed for tool use. Lucas: Right. One approach is to redesign the tool itself — make scissors with a standardized attachment point and a built-in actuator, so the robot doesn't have to mimic the human grasp. That's what you see in some automated surgical systems. Luna: So instead of teaching the robot to use human scissors, we give it robot scissors. Lucas: Exactly. The other path is to improve the hand — using underactuated grippers that conform to the handle shape, or adding tactile sensors that detect slip in real time. Some groups are using soft robotic fingers that can adapt to the taper of plastic handles. Luna: But even with those, the control problem remains. You need to model the dynamics of the pivot and the cutting action simultaneously. Lucas: Right. And that's where reinforcement learning comes in. Researchers at Carnegie Mellon recently trained a robot arm to cut cloth with scissors by practicing thousands of times in simulation. The success rate was around seventy percent on straight cuts. Luna: Seventy percent for straight cuts on cloth — that's promising, but not ready for the factory. Lucas: No, not yet. But it shows that with enough data and the right simulation environment, we can bridge the gap. The challenge is sim to real transfer — what works in simulation often fails on real scissors because of friction, slop in the pivot, lighting changes. Luna: So maybe a hybrid approach: use a purpose-built tool but with a gripper that mimics human dexterity? Lucas: That's where I think the field is heading. Not trying to copy human hands exactly, but designing tools that are robot-compatible and hands that are tool-compatible. It's a co-optimization problem. Luna: And when it works, the applications are huge — from automated garment manufacturing to robot-assisted surgery, even household service robots. Lucas: Right. But for now, if you hand a robot a pair of scissors, it's going to have a bad day. And that's actually a good reminder of how much subtle intelligence goes into even the simplest human actions. Luna: So next time you cut a coupon, give yourself some credit. Lucas: Exactly. And maybe spare a thought for the robot that will keep messing that up for another few years.