Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Hands Still Cant Handle a Ziploc Bag
Transcript
- Lucas: Luna, I tried to open a ziploc bag this morning with my eyes closed. It took me about two seconds. A robot with a ten thousand dollar gripper and a vision system would fail that test nine times out of ten. Luna: Wait — the same robot that can assemble a gearbox can't open a sandwich bag? Lucas: Exactly. And that's what I want to talk about today. Because the ziploc bag is one of those objects that reveals a huge blind spot in modern robotics. It's thin, flexible, transparent, static-prone, and the opening force changes depending on where you pull. Luna: And we've already covered soft objects, wet objects, cables — this feels like the next frontier of annoying everyday things. Lucas: Right. So let's break down why it's hard. First, the material itself. A typical ziploc bag is made of low-density polyethylene, about 0.04 millimeters thick. That's thinner than a human hair. Most robot grippers are designed for rigid objects with known geometry. A vacuum gripper can pick up a sheet of paper, but a ziploc bag — because it's so light and flexible — tends to wrinkle or fold over itself, breaking the suction seal. Luna: And even if you do get suction, the bag often just sticks to the gripper in a crumpled mess instead of opening neatly. Lucas: Precisely. The second issue is transparency. Most ziploc bags are clear, so standard depth cameras struggle to detect the edges and the opening slit. The bag basically disappears against a white table or a stainless steel countertop. Luna: So the robot can't even see where to grab. Lucas: Correct. And even if you solve vision, the third problem is the physics of opening. To open a ziploc bag, you need to create a small gap between the two seal ridges — usually by pinching one side and pulling. That requires applying force in a very specific direction, and the bag offers almost no resistance until the seal pops. Hard to control with a rigid gripper. Luna: I've seen some research on using electrostatic adhesion for thin films. Is that relevant here? Lucas: Very relevant. A team at MIT's CSAIL published a paper in 2024 — actually, updated with new results earlier this year — where they used electroadhesion to pick up single sheets of polyethylene. By applying a voltage to a conductive pad, they create an electric field that polarizes the film and holds it without any mechanical pinch. They managed to pick up and lift a ziploc bag by one corner, then shake it open. Luna: But shaking it open is not the same as opening the seal and keeping it open while you put something inside. Lucas: That's the follow-up challenge. The CSAIL group is now working on a two-gripper system — one electroadhesive pad to hold the top edge, another to hold the bottom, and then a peel motion. But the timing has to be perfect. If you peel too fast, the bag tears. Too slow, it won't open at all. Luna: Is there any commercial system that can do this today? Lucas: Not that I know of. There are food-packaging robots that open pre-cut bags, but those bags are not resealable — they're heat-sealed on three sides and the robot just tears a perforated line. A resealable ziploc is fundamentally different because the seal is designed to be reusable, which means it's stronger and requires a precise peel force. Luna: So this is still a research problem. Lucas: Mostly. But there are startups tackling it. One I've been watching is a company called Dexterity — they focus on warehouse picking of mixed items. They've developed a gripper with embedded tactile sensors that can detect the edge of a bag using capacitive sensing, not vision. They demoed picking up a ziploc of almonds last year at a trade show, but it was a controlled environment with specific lighting. Luna: So we're not at the point where a robot can walk up to your kitchen counter and open any plastic bag. Lucas: Not even close. And that's actually the deeper point here. The ziploc bag is a perfect emblem of a broader problem: the gap between human-level dexterity and robotic manipulation. Humans have about twenty-two degrees of freedom in the hand, plus thousands of tactile receptors per square centimeter of fingertip skin. A robot gripper might have two degrees of freedom and a handful of force sensors. Luna: We've talked about folding t-shirts, threading needles, opening jars — and now bags. It seems like any task involving a compliant, deformable object is a nightmare for robots. Lucas: Exactly. And the reason is that these objects don't obey rigid-body dynamics. Their geometry changes as you interact with them. You can't model them with a fixed CAD file. You need real-time sensory feedback and adaptive control. And that requires hardware that is still very expensive. Luna: Which brings us to the economics. How much would a robot hand that could reliably open a ziploc bag cost today? Lucas: If you use a high-end dexterous hand like the Shadow Hand — that's about one hundred twenty thousand dollars for the hand alone, plus the arm, plus the vision system. You're looking at a quarter of a million dollars for a system that still fails on a bad day. Compare that to a human minimum-wage worker who can do it instantly. The ROI isn't there. Luna: Unless you're in a sterile environment where you don't want human hands touching the product — like medical packaging. Lucas: Bingo. That's where the economic case flips. In pharmaceutical cleanrooms, every human touch introduces contamination risk. A robot that can open a sterile pouch and dose medication inside without touching it — that's worth the investment. A company called OmniCell has been working on robotic pharmacy systems, but they still rely on pre-opened pouches. Luna: So maybe the killer app for ziploc-opening robots isn't your kitchen — it's the hospital pharmacy. Lucas: Right. Or food processing plants where workers have to open thousands of bags of ingredients per shift. That's repetitive strain injury territory. If a robot could do it with even eighty percent reliability, it would still be worth it. Luna: And I guess that's the practical lens we should keep on this. It's easy to scoff at a robot that can't open a bag of chips, but the underlying technology — sensing thin flexible materials, applying variable force, handling transparency — that has real industrial value. Lucas: Absolutely. And look, one of the reasons we can dig into niche problems like this on a daily basis is that this show is listener-supported. No ads, no sponsors. We're funded by a small group of people who chip in through buy me a coffee dot com slash fexingo. Luna: It's true. And that support lets us spend time on topics that might not get coverage on a bigger ad-driven show. So if you're getting value from these deep dives, consider tossing a couple bucks in the virtual jar. Lucas: Okay, back to the bag. I want to mention one more approach that I find really clever. Researchers at the University of Tokyo published a paper last year using a pair of grippers with a small air nozzle. They would first blow air into the bag to inflate it slightly, then grab the top and pull. The inflation creates a gap that makes the seal easier to break. Luna: So you're using fluid dynamics to solve the mechanical problem. Lucas: Exactly. And it worked in about seventy percent of trials with standard ziploc bags. The failure cases were mostly when the bag had a pinhole leak or when the seal was too tight. So it's not perfect, but it's a creative workaround. Luna: What about machine learning? Could you train a model to find the optimal opening strategy for any bag? Lucas: People are trying. One challenge is that the reward function is hard to define. How do you score a partially open bag? Another is that the physical simulation of thin-film materials is notoriously difficult. Simulating a ziploc bag with realistic friction, electrostatic forces, and tearing is computationally expensive. Luna: So we're back to the fundamental problem: the real world is messy. Lucas: It is. And that's what makes this field so interesting. Every 'simple' human task you break down reveals layers of complexity. The ziploc bag is a microcosm of why general-purpose robotics is still decades away. Luna: But at least we know what the next micro challenge is. Lucas: Right. Because after the ziploc, there's the bread bag twist tie. And after that, the produce netting. And then the vacuum-sealed coffee bag. Each one has its own failure mode. Luna: So maybe in episode 60 we'll be talking about robots that can open a bag of frozen peas. Lucas: I'd bet on it. And we'll keep following the progress. For now, I'm Lucas. Luna: And I'm Luna. Thanks for listening.