Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Cant Handle a Deflated Ball
Transcript
- Lucas: You know that feeling when you try to pick up a deflated beach ball? The thing just collapses in your hand. Luna: Oh, absolutely. It's floppy and unpredictable. And I'm guessing that's exactly the kind of problem robot arms still can't solve. Lucas: Right. We've done a bunch of episodes on why robots can't handle things like a wet towel or a single grain of rice. But today I want to talk about a specific object: a deflated ball. Not a balloon, not a rigid sphere — a soft, partially collapsed ball. Luna: Why that object in particular? Lucas: Because it sits right at the intersection of two challenges: it's deformable, so its shape changes when you touch it, but it also has internal pressure that shifts as you squeeze. A team at Disney Research actually published a paper on this in 2023, and it's a great case study. Luna: Disney Research — they're the ones who do all those animatronics and soft robotics projects, right? Lucas: Exactly. They wanted to build a robotic hand that could catch and throw a ball, but they realized that a standard rigid gripper couldn't handle a ball that's even slightly deflated. So they set up an experiment. Luna: What did they do? Lucas: They took a standard industrial robot arm with a two-finger gripper and tried to pick up a partially deflated volleyball. The gripper would close, and the ball would just squish out sideways. The fingers couldn't apply enough force without damaging the ball, and the lack of a fixed geometry meant the grip was unstable. Luna: So the robot couldn't even hold it, let alone throw it. Lucas: Right. And this isn't just a toy problem. Think about warehouse automation: boxes that are slightly crushed, bags of flour, or soft pouches of liquid. These are all deformable objects that current robot arms handle poorly. Luna: What's the fundamental issue? Is it sensing, or control, or both? Lucas: Both, but let's start with sensing. Most industrial robot arms use joint torque sensors or force-torque sensors at the wrist. Those measure the total force applied, but they don't tell you the pressure distribution across the gripper's fingers. So when you grab a deflated ball, the sensor might say 'I'm applying 5 Newtons,' but it doesn't know that the ball is deforming and the contact area is changing. Luna: So the robot is basically blind to how the object is actually behaving in its grip. Lucas: Exactly. Human hands have thousands of tactile receptors. We can feel a ball starting to slip and adjust our grip in milliseconds. Robots don't have that. The Disney team tried adding pressure-sensitive rubber to the gripper fingers, and it helped a little, but it still couldn't adapt fast enough. Luna: What about control algorithms? Could you model the deformation and predict how the ball will behave? Lucas: That's the holy grail. They tried using a physics simulation to predict the ball's shape as it's squeezed. But a deflated ball has a non-linear, hysteretic behavior — meaning how it deforms depends on its previous state. If you've already squeezed it once, it doesn't bounce back to the same shape. So the model quickly diverges from reality. Luna: So even with a simulation, you can't keep up. Lucas: Not reliably. And that's why a deflated ball is such a good test case: it's simple in concept but exposes every limitation of current hardware and software. Luna: Before we go further, a quick honest thing — a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that support is what keeps this show ad-free and lets us dive into niche stuff like robot grippers. So if you get value from these deep dives, that's the button. Lucas: Yeah, we don't run ads, so it's listener-funded. We really appreciate it. Okay, back to the ball. Luna: So, does any robot arm out there actually handle a deflated ball well? Lucas: Not with standard hardware. But there are some experimental approaches. One is using a soft gripper — think of a rubbery, inflatable hand that conforms to the object. The Octopus Gripper from Festo is a good example. It's designed to handle irregular shapes. But it's slow and not precise. Luna: How slow? Lucas: On the order of seconds to grasp, compared to a rigid gripper that can close in a fraction of a second. In a warehouse, that time adds up. Luna: What about using suction? Like a vacuum gripper? Lucas: Suction works for flat or smooth surfaces, but a deflated ball has a curved, porous surface. The vacuum seal is poor. And if the ball has any wrinkles, the suction cup leaks. So that's not reliable either. Luna: So we're stuck. Any promising research beyond Disney's? Lucas: MIT's CSAIL lab has a project called 'GelSight' — a tactile sensor that uses a camera to image the deformation of a soft gel pad. It gives high-resolution pressure maps. They've used it to grasp a deflated ball with a custom gripper. But it's still lab-scale, not industrial. Luna: So the deflated ball is basically a benchmark for how far we are from human-like dexterity. Lucas: Exactly. It's a simple object that reveals a huge gap. And until we solve that, robots will keep struggling with anything that isn't rigid and predictable. Luna: What's the takeaway for someone in robotics or automation? Lucas: If you're designing a robot workcell for a task that involves soft, deformable objects, don't assume a standard gripper will work. You need to invest in tactile sensing and adaptive control. And even then, set expectations low. The deflated ball is still a frontier problem. Luna: It's humbling, isn't it? That a piece of soft rubber can outsmart a million-dollar robot arm. Lucas: It really is. And that's what keeps this field so interesting. Every time we think we've cracked manipulation, the deflated ball reminds us how much we don't know.