Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Hands Can't Handle a Ziploc Bag
Transcript
- Lucas: So we've spent over a hundred episodes watching robot arms fail at things toddlers can do — peeling potatoes, tying knots, threading needles. This week I want to talk about one of the most mundane, frustrating failures I've seen in a while. Luna: Let me guess. It involves something floppy and thin. Lucas: Exactly. A Ziploc bag. Not even a full bag — just the act of opening one. If you've ever been in a lab or a kitchen and watched a robot try to separate the two sides of a resealable plastic bag, you know it's a catastrophe. The grippers squeeze, the bag crumples, the seal stays shut, and you end up with a tangled plastic mess. Luna: I've seen videos. It's like watching a robot fight a ghost. So what makes a Ziploc bag so hard? Lucas: Two main problems. First, the material itself. Polyethylene film is thin, flexible, and has very low friction. When a typical parallel-jaw gripper closes on it, the film just slides out — there's no purchase. Second, opening the bag requires applying force in two opposing directions to separate the seal, but you also need to keep the opening from collapsing. It's a multi-contact manipulation problem that most industrial grippers aren't designed for. Luna: Right, because a robot that's great at picking up a metal part from a bin has no concept of 'gentle peel.' Lucas: None. And that's why a team at MIT's CSAIL recently published a paper specifically on this. They tested over 200 different opening strategies using a variety of grippers — suction, pinch, tweezer-style, even electrostatic adhesion. And the success rate across the board was surprisingly low. Luna: What was the best strategy they found? Lucas: A combination of suction and a peel motion. So you use a suction cup to lift one layer of the bag, creating a small air gap. Then a second gripper pinches the other layer and pulls it away at an angle, like you'd peel a sticker off a surface. That gave them about a 78 percent success rate in lab conditions. But if the bag had any static cling or moisture? That dropped to under 50 percent. Luna: So even the best solution is fragile. What about those soft robotic grippers that are supposed to be gentle? Lucas: You'd think soft grippers would be the answer, and they are better — but they introduce their own problems. Soft grippers made from silicone or rubber conform to the bag's shape, which helps with grip. But they also deform unpredictably. The bag can wrinkle and fold in ways that make it harder to separate the layers. And if the gripper is too soft, it doesn't generate enough force to break the seal. So you end up with a trade-off: too soft, can't open; too hard, crushes the bag. Luna: There's a Goldilocks problem in material science. And this isn't just a kitchen curiosity, right? There are real applications where opening a plastic bag matters. Lucas: Huge ones. Think about lab automation. Pharmaceutical labs handle thousands of sterile sample bags every day. Currently, humans open them manually because robots can't reliably do it. Same with food processing — packing lettuce or frozen vegetables often involves opening a plastic liner inside a box. And in logistics, there's a growing need for robots to open poly mailers for returns processing. Amazon alone processes millions of returns that come in plastic bags. If a robot can't open the bag, a human has to. Luna: That's a lot of labor hours that could be automated. So what's the state of the art now? Are any companies solving this? Lucas: A few startups are working on it. One company called Soft Robotics Inc. has a gripper that uses a combination of suction and a rolling motion — it basically rolls the bag open like you'd unroll a sleeve. They claim about 85 percent reliability in controlled tests. Another approach uses electrostatic adhesion: a charged pad attracts one layer of the bag, then lifts it while a second gripper peels the other side. That's still in research. Luna: But none of these are commercial yet, right? I haven't seen a robot in a grocery store opening bags for produce. Lucas: Not yet. The big barrier is perception. Robots need to see the bag's seal line and the orientation of the layers, and that's hard with transparent or translucent plastic. Even with lidar or depth cameras, the thin film can be nearly invisible. So you need either specialized lighting or a tactile sensor that can feel the seam. And that drives up cost. Luna: So we're at the point where the hardware is getting there, but the sensing still lags. And this is where the listener might wonder — why does this matter for business? Who's investing? Lucas: Good question. The market for flexible packaging automation is estimated at about $4 billion globally, and it's growing at roughly 8 percent a year. Companies like Sealed Air and Procter & Gamble have internal R&D projects on this. And venture capital is starting to flow — there were at least three early-stage rounds last year for gripper companies specifically targeting thin-film manipulation. Luna: So there's a real economic incentive. But I'm curious — you mentioned MIT's CSAIL paper tested 200 strategies. What were the most surprising failures? Lucas: One that stood out: using a vacuum gripper with a single suction cup. The idea was to suck one layer up, then slide the other layer off. But the vacuum often pulled both layers together if the bag was sealed too tightly. And when it did separate them, the suction cup sometimes collapsed the bag entirely, making it impossible to access the opening. Also, any static cling — which is common with polyethylene — caused the layers to stick back together immediately. Luna: So static electricity is a real enemy here. That's something humans don't even think about — we just rub the bag open. Lucas: Exactly. Humans have this intuitive understanding of friction, adhesion, and static. We adjust our grip angle, we slide our fingers to create a gap, we use our fingernails to separate the seal. Robots have none of that. And replicating a fingernail — a rigid, thin edge with precise force control — is surprisingly hard. Some researchers have tried adding a 'fingernail' to the gripper, but it tends to puncture the bag. Luna: So what's the most promising path forward? Is it better hardware, better software, or a combination? Lucas: I think it's a combination, but the software side might be more important than people realize. The MIT team used reinforcement learning to let the robot experiment with different opening motions. Over thousands of trials, it learned a sequence that worked well: first, press down lightly to create a crease near the seal, then use a suction cup to lift one side, then slide a second gripper under the other side and pull. That sequence isn't obvious to a human programmer. It emerged from the data. Luna: So the robot basically taught itself to open a bag. That's a different approach than designing a special-purpose gripper. Lucas: Right. And it suggests that general-purpose robot hands — like the ones being developed by companies like Shadow Robot or the DLR — could eventually learn to do this with the right training. But we're still years away from a robot that can reliably open any Ziploc bag in any condition. Luna: Speaking of learning from data — I think a lot of our listeners enjoy diving into these technical rabbit holes with us. If you're finding these deep dives valuable, you can support the show at buy me a coffee dot com slash fexingo. It keeps the podcast ad-free and helps us keep exploring topics like this one. Lucas: Yeah, and we really appreciate it. It's what allows us to spend time on these weird, specific problems instead of just covering the same news stories as everyone else. Luna: So back to the bag problem — where do you see this heading in the next five years? Lucas: I think we'll see specialized bag-opening stations in controlled environments first — like in pharmaceutical labs where the bag size and material are standardized. Those could become reliable within two to three years. For general-purpose bag opening in messy environments like grocery stores or warehouses, probably five to seven years. The key will be integrating tactile sensing with vision, and maybe using active cooling to reduce static cling. Luna: Active cooling? That's a new one. How would that work? Lucas: Some researchers have found that cooling the bag reduces static charge. So if the gripper has a small Peltier element, it can chill the film just before attempting to open it. That's still experimental, but it shows how creative these solutions have to be. Luna: It also shows how far we are from robots that can handle the mundane tasks we take for granted. Lucas: Exactly. And that's what makes this field so interesting to follow. Every time a robot fails at something simple, we learn something about how complex human manipulation really is. Luna: Alright, next episode I want to see if a robot can open a bag of chips without crushing them all. Lucas: That's a whole other level of difficulty. Let's put it on the list.