Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Still Cant Handle a Pill
Transcript
- Lucas: Alright, Luna, here's a scenario. You walk into a pharmacy, hand the pharmacist a prescription, and instead of a person counting out pills, a robotic arm picks each one from a blister pack and drops it into a bottle. Luna: I've seen those automated dispensing machines in hospitals. They mostly use a vibratory feeder or a bulk hopper. But picking a single pill from a blister—that's a different beast. Lucas: Exactly. And it turns out, that's the unsolved problem we're drilling into today. Because a standard industrial robot arm, the kind that welds car frames or sorts packages, cannot reliably pick a single pill from a blister pack. It's embarrassingly bad at it. Luna: So we're doing another 'why robots can't handle X' episode. What makes the pill special? Lucas: Three things: geometry, surface properties, and vision. First, pills are small, round, and often coated with a low-friction film—think of the shiny coating on an ibuprofen tablet. That coating is designed to make it easy for humans to swallow, but it makes it nearly impossible for a robot gripper to get a secure hold. Luna: Right, because the gripper either slides off or applies too much pressure and crushes the pill. I remember reading a study from MIT back in 2025—they tested several gripper types on common over-the-counter pills. Lucas: That's the one I'm referencing. The MIT team used a standard parallel-jaw gripper, a three-finger gripper, and a soft pneumatic gripper. They tried to pick up ibuprofen tablets, each about 12 millimeters in diameter, from a flat surface. The success rate for the parallel-jaw gripper was below thirty percent. Luna: Thirty percent? That's terrible. A human can do it blindfolded. Lucas: And with one hand. But the robot struggled because the pill's low-friction coating caused it to slip out of the grip. The soft gripper did better—about sixty-five percent success—but it still failed when the pill was slightly rotated or if the blister pack had any residual foil edge. Luna: So the material science is part of it. But also the environment. A blister pack is shiny, reflective, and the pill sits in a cavity. For computer vision, that's a nightmare. Lucas: That's the second challenge. The blister pack's foil backing is highly reflective. A standard RGB camera sees glare and reflections that confuse segmentation algorithms. Even with structured light or depth sensors, the pill can appear to merge with the background. The MIT team had to use a combination of infrared illumination and a polarizing filter to get a clean image. Luna: And even then, the gripper has to approach at exactly the right angle. If it comes in too steep, it might hit the blister 'bubble' wall. Too shallow, and it misses the pill entirely. Lucas: Right. The third challenge is precision. The gap between the pill and the blister cavity is often less than a millimeter. So the robot needs to know the exact three-dimensional position and orientation of each pill. If the blister pack shifts even slightly on the conveyor, the whole pick fails. Luna: This is exactly the kind of problem that makes you appreciate how amazing human hands are. We just reach in and grab it, adjusting force and angle without thinking. Lucas: And we do it with incredibly low error rates. In the same MIT study, human subjects achieved a ninety-nine percent success rate picking pills from blisters. The robots, even with optimized grippers, maxed out at around seventy-five percent in ideal conditions. Luna: So why does this matter? Pharmacy automation is a huge market. If you can automate the dispensing of pills, you reduce errors and free up pharmacists for more clinical work. Lucas: Exactly. There's a company called MedAvail that has automated pharmacy kiosks, but they use pre-filled cassettes—they don't pick from blisters. The holy grail is a robot that can handle any pill from any blister, because that's how most medications are packaged in many countries. Luna: And with an aging population, demand for accurate, high-volume dispensing is only going up. Lucas: So what's being done about it? The most promising approach right now is suction-based picking. Instead of grabbing the pill, you use a small vacuum nozzle to lift it. But even that has issues—if the pill has a hole in the middle, like a donut-shaped tablet, the suction can't hold it. Luna: Or if the pill is porous, like some supplements, the vacuum leaks. So you need to sense the pill's surface properties in real time. Lucas: That's where sensor fusion comes in. Researchers at Carnegie Mellon are working on a gripper that combines tactile sensors, force feedback, and a small camera right at the fingertip. It can detect if the pill is slipping and adjust grip pressure in milliseconds. Luna: But those systems are still in the lab. The MIT study noted that the best performing gripper still failed ten percent of the time when the pill had any surface moisture—like from humidity. Lucas: Which is a real problem because pharmacies aren't climate-controlled clean rooms. A pill can have condensation from being in a cold storage area. That changes the coefficient of friction dramatically. Luna: So we're looking at a combination of better gripper materials, smarter vision, and maybe even redesigning the blister pack itself to make it robot-friendly. Lucas: Some pharmaceutical companies are exploring that—making the foil backing slightly matte or adding a small tab that the robot can grip. But that requires regulatory approval, which takes years. Luna: And in the meantime, the robots keep failing on standard blisters. It's a good reminder that even in 2026, some of the most mundane tasks are still beyond the reach of automation. Lucas: And speaking of keeping things going—if today's tech conversation gave you something usable, we want to mention that this podcast is ad-free by design. We deliberately don't run ads on these. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Luna: Yeah, listener support is what keeps it that way. No pressure, just if you find value in the deep dives. Lucas: Exactly. Back to pills—so where do we go from here? One emerging idea is to use a soft gripper with a patterned surface, like micro-suction cups inspired by gecko feet. That could provide grip without crushing. Luna: Gecko-inspired grippers have been used for handling delicate objects like silicon wafers. Adapting them to pills makes sense. Lucas: But they're expensive and not yet scalable. Another approach is to use a jet of air to levitate the pill slightly before suction picking. That reduces the need for precise initial positioning. Luna: So it's a multi-pronged problem. Vision, grip, and environment all have to be solved together. Lucas: And that's what makes it interesting. It's not just about building a better gripper—it's about building a system that can perceive and adapt. The same technology that would let a robot pick a pill could also let it pick a slippery battery or a delicate electronic component. Luna: So the payoff is bigger than just pharmacy automation. Lucas: Much bigger. It's about achieving the kind of general dexterity that humans take for granted. And right now, a simple pill is one of the best benchmarks we have. Luna: Alright, so next time I pop an ibuprofen, I'll appreciate that no robot could have gotten it out of the pack for me. Lucas: Not yet. But give it a few more years. The research is accelerating, and I think we'll see a commercial solution within the next three to five years. Luna: I hope so. My pharmacist would appreciate it. Lucas: And that's the show for today. We'll be back next time with another look at the frontier of robotics.