Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Handle a Bowling Pin
Transcript
- Lucas: So I was at a robotics demo day last month, and one of the researchers had set up what looked like a pretty simple challenge: a robot arm next to a standard ten-pin bowling pin, just sitting there on a table. The arm had to pick it up, rotate it 180 degrees, and set it back down. Luna: Okay — that sounds like something a six-axis arm should be able to do in its sleep. Lucas: That's what I thought. But the arm failed. Over and over. It would try to pinch the pin around the midsection, slide, regrasp, knock it over. The researcher told me they've been working on this specific task for eight months. Luna: Eight months on a bowling pin. That's both absurd and completely believable. Lucas: Right. And that's what we're going to unpack today — why a bowling pin is a genuinely hard object for a robot, and what that tells us about the state of manipulation. The pin looks simple, but it has three properties that break most grippers. Luna: Let's hear them. Lucas: First: its center of mass is way below its geometric center. A bowling pin is basically a fat cylinder that's narrow at the top and wide at the bottom, but the weight is all in the base — about two-thirds of the mass is in the bottom third of the pin. So when a gripper grabs it near the neck, the pin wants to tilt and swing like an inverted pendulum. Luna: And most robot grippers are designed for objects where the center of mass is roughly in the middle. They assume symmetry. Lucas: Exactly. The second problem is surface friction — or rather, the lack of it. Bowling pins are coated with a high-gloss polyester resin. It's incredibly slick. So parallel-jaw grippers with hard plastic or metal fingers just slide off. You need high-friction pads, but even those struggle if the pin isn't perfectly dry. Luna: And the third problem? Lucas: The pin has to be reoriented in mid-air. In a real bowling alley, pins are set by a machine that drops them into a fixed pattern, so orientation is predetermined. But in a warehouse scenario — say a robot has to pick a pin from a bin and place it onto a conveyor — it needs to sense the pin's current pose and then rotate it to the correct orientation, all while maintaining a stable grip on a slippery, off-balance object. Luna: That reorientation step is where the failures really pile up. I've seen videos where the arm lifts the pin, tries to rotate its wrist, and the pin just flips out of the grasp. Lucas: Exactly. And this isn't some fringe lab curiosity. The bowling pin problem maps directly to a class of objects that warehouses handle all the time: bottles with narrow necks and heavy bases, automotive components with irregular mass distribution, even some power tools. If you can solve the bowling pin, you've cracked a big chunk of the general grasping problem. Luna: So who's working on it? Lucas: There's a team at the University of California, Berkeley, that published a paper last year on what they call 'dynamic re-grasping' — the robot basically bounces the object between its fingers to shift the grip point. They tested it on wine bottles, which have a similar mass imbalance. Their success rate went from about 40 percent to 85 percent. Luna: Eighty-five percent is good, but still not production-ready. In a warehouse, you'd want something like 99.9 percent. Lucas: Right. And the fix isn't just software. A company called Festo has been working on a gripper that uses a flexible membrane filled with granular material — it conforms to the object's shape and then stiffens by applying vacuum. It's called the FinGripper. It can handle a bowling pin, but it's slow — the vacuum takes about a second to engage. Luna: Speed versus reliability. That's the trade-off that keeps coming up. Lucas: It really is. And that trade-off is especially sharp right now because the warehouse robotics market is exploding. According to a report from MarketsandMarkets, the global warehouse robotics market is projected to reach $30 billion by 2030. But the bottleneck isn't the mobile base or the vision system — it's the gripper. The end of arm tooling that actually touches the product. Luna: So the bowling pin is almost like a benchmark. If your gripper can handle that, it can probably handle a lot of the messy real-world objects that currently require human hands. Lucas: That's exactly how the research community is using it. There's actually a standardized test called the 'YCB Object Set' — yale cmu berkeley — that includes a bowling pin among 77 objects designed to probe different grasping challenges. And year after year, the bowling pin remains one of the lowest-scoring objects across all submitted algorithms. Luna: You know, this reminds me of a conversation we had a few episodes back about robot arms and wet paper towels. The physical world just doesn't cooperate with the tidy assumptions that work in simulation. Lucas: That's the thread running through so many of these episodes. And look, this podcast is possible because listeners support us to keep digging into these stories. If today's tech conversation gave you something usable, a couple of dollars a month is genuinely what keeps these going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. Luna: Yeah, it really does make a difference. We don't run ads, so listener support is what keeps this ad-free and focused on the real engineering challenges. Lucas: Alright. Back to the pin. One approach that's gaining traction is using multiple grippers on the same arm — a kind of tool-changer. The robot picks up the pin with a soft gripper for the initial lift, then transfers it to a rigid gripper for the reorientation. But that adds cycle time and mechanical complexity. Luna: Is anyone using suction? A vacuum cup on the top of the pin? Lucas: That works if the top surface is smooth and clean, but on a used bowling pin, the top is often worn or has a small hole from the pin-setting machine. Plus, suction is terrible for objects with holes or irregular tops. So it's not a general solution. Luna: So what's the most promising direction you've seen? Lucas: I'd put my money on tactile sensing combined with learned control policies. A team at MIT recently demonstrated a gripper with high-resolution touch sensors that can detect slip before it happens — they measure shear forces at the contact points in real time and adjust grip force accordingly. On a bowling pin, they achieved a 90 percent success rate on the pick and place task, including reorientation. Luna: Ninety percent. That's getting close. Lucas: Close, but still not there. And the interesting thing is, the failures weren't due to the gripper itself — they were due to the pin's surface condition. If the pin had even a trace of lane oil from the bowling alley, the coefficient of friction dropped by about 30 percent, and the gripper couldn't compensate fast enough. Luna: So the environment matters as much as the object. Lucas: Exactly. And that's the deeper lesson: robust manipulation isn't just about the robot or the gripper — it's about understanding the full context. The lighting, the surface texture, the presence of contaminants, the variability in object shape from one pin to the next. Even though bowling pins are manufactured to tight tolerances, they do wear unevenly. Luna: So a robot that works on a brand-new pin might fail on one that's been used for fifty games. Lucas: Exactly. And that's the gap between a lab demo and a production system. In a real warehouse, the objects are never pristine. They come in all conditions. That's why companies like RightHand Robotics are building grippers that combine multiple sensing modalities — vision, force, and tactile — and then use a learned model to decide, on the fly, which strategy to use. Luna: It sounds like the bowling pin is a perfect test case for that kind of multi-modal approach. Lucas: It really is. And I think we'll see a solution within the next two to three years that handles it reliably. Once that happens, a whole class of warehouse picking tasks that currently require human hands will suddenly become automatable. Luna: And then maybe the robot can finally rack the pins for us too. Lucas: That's actually a much harder problem — you have to place them in a precise triangular formation, and the pin spacing is tight. But that's a topic for another episode. For now, I'm just glad we understand why that simple-looking cylinder is such a tough customer. Luna: It really makes you appreciate what human hands do effortlessly. Lucas: Until next time.