--- title: "A Robot's First Real Job: Learning Every Box on the Pharmacy Shelf" date: 2026-11-05 category: Humanoids & Robotics site: NeuroAI canonical: https://neuroai.site/a/na-pharmacy-robot-16-1105 language: en --- # A Robot's First Real Job: Learning Every Box on the Pharmacy Shelf > In a corner of the "Light of the Internet" expo in Wuzhen, a space has been fitted out as a pharmacy. Shelves, counter and pickup window are all there. The only difference is that the person behind In a corner of the "Light of the Internet" expo in Wuzhen, a space has been fitted out as a pharmacy. Shelves, counter and pickup window are all there. The only difference is that the person behind the counter is not a person. It is a machine on wheels. Visitors place an order on a phone. The robot moves between the shelves, finds the medicine, reaches out, takes it, and delivers it to the pickup window. The whole process is compressed into a few square meters. This is one booth at the 2026 World Internet Conference. It is not flashy. There is no music, no lighting, no narrator. But it may be the most concrete job robots have right now, and the easiest to underestimate. ## A Mock Pharmacy According to Xinhua, in this mock retail pharmacy, a robot running Ant Group's LingBot model can nimbly grasp medicines in different packaging. Boxes, bottles and soft bags all have to be picked up. One line from the floor has been quoted by several outlets: "This robot has already been put into use in multiple pharmacies in Shanghai's Pudong district." The speaker was Duan Xiaoyue, a market operations specialist at Ant Group's Lingbo unit, standing beside the booth. Note the word "multiple." No store count, no coverage ratio, no number of orders handled per day. That is as far as the report goes. Anything beyond it would be invention, and invented numbers are the easiest thing to get wrong. The same report contains another figure: the LingBot series already supports robots from 17 brands and more than 20 different configurations. That sentence is more worth pondering than "it can pick up medicine," because it is about generality, not a single trick. ## Why Medicine Boxes Are Hard To an outsider, picking up a box sounds trivial. Put a pile of medicines in front of you and the problems start immediately: hardly any two boxes are the same size, material or shape, and even different batches of the same drug differ. There are hard cartons, soft bags, plastic bottles and foil blister packs. Some surfaces are smooth and reflective, some corners dig into the hand, and some are stacked so that pulling one out knocks over a row. You need the other hand to steady them, or you have to restack the whole shelf. Human hands are dense with touch receptors. Squeeze too hard and you crush the box; too lightly and it slips. A robot has no sense of touch, only vision and force sensors, and it learns by trial, where every error is a redo. Position is harder still. A box is never exactly in the middle of its slot. The previous person pushed it askew, the neighboring box squeezed it, it fell over and landed face down with the label turned the wrong way, unreadable at a glance. So picking medicine is not a test of strength. It is a test of whether the chain of see, judge, reach and adjust connects in one go. Break the rhythm once and the box drops. Then you start over, and it may break, and if it breaks you pay. ## In a Pharmacy, Mistakes Matter Pick the wrong carton of tissues in a warehouse and the loss is a few yuan. Pick the wrong box of medicine in a pharmacy and it is a different matter. Behind the box is a specific person, with a name, a medical history, and other drugs they are already taking. That is why one line from the on-site introduction matters: before medicine leaves the pharmacy it still has to be checked by a pharmacist. The robot only fetches it. The last gate is still a human, a point repeated several times at the booth. That line blocks two fantasies. One is the "unmanned pharmacy." The other is "pharmacists will lose their jobs." In this scheme, at least, neither is happening, and neither should be written as if it were. A writer has to hold that line. What the robot takes over is the walking, the searching and the bending: the time-consuming parts. It covers a lot of ground in a day. The time pharmacists free up goes into checking prescriptions and explaining how to take the medicine, things a machine cannot yet do. The division of labor is not glamorous, but it is realistic. Medicines are heavily regulated goods. The final step requires a signature. Whoever signs is responsible, and if something goes wrong, there has to be a person to find. That is the last door robots cannot walk through. ## Why Pharmacy First The two settings most often promised for humanoid robots are factories and homes. One is too rigid and the other too messy. Neither is easy: one requires shutting down a line for retrofitting, the other means facing endless surprises. Neither can be rushed. A factory line is designed around a takt time. Put a robot in and you often have to rebuild the whole line. A home is full of surprises: a pair of shoes on the floor today, a cardboard box tomorrow, a cat running around the day after. A pharmacy sits in between. It has shelves, a fixed route and clear tasks, but its objects are messy enough to test a pair of hands, and not so messy that nothing can be organized. That balance is hard to strike. On top of that, pharmacy orders are fragmented. A few boxes per order, one order after another, and people order at night too. This is the most grinding kind of work, and the best suited to something that does not tire, does not get annoyed and does not misread an order. So the pharmacy is not the destination. It is a training ground. The grasping, recognition and verification learned here can move to convenience stores, warehouses and kitchens, even to the parcel station downstairs. The order is probably much the same. ## The Model Business Behind One Machine What Ant Group's Lingbo really wants to sell may not be the robot but the model called LingBot. The robot is only a shell. The brain is the goods. Shells can be swapped; brains are expensive to swap, because every swap means retraining. Supporting 17 brands and more than 20 configurations says it plainly: buy a robot from anyone, as long as it runs my brain, and I can put it to work. Never mind how many axes its arm has or how many fingers its hand has. This is a familiar playbook. Hardware blooms, then consolidates around a few general-purpose brains. Whoever's brain is installed most widely gets a chance to set the standard, and once the standard is set, everyone else can only follow. But the road has a weak point. To fit into someone else's machine, you have to adapt to their hands, joints and sensors. Every brand needs retuning, and a bad job damages both reputations. Neither side can afford that. So the sentence "supports more than 20 configurations" is both an achievement and a workload. It is not a number. It is more than twenty sets of tuning records, hundreds of on-site trials, countless reworks, and a pile of logs nobody reads. ## The Least Glamorous Setting This year's expo drew 671 companies from 54 countries and regions, showing more than 1,000 new technologies and products. On November 2, 20 achievements received the World Internet Conference Leading Science and Technology Award, chosen from more than 430 submissions from 43 countries and regions. Among all that, a robot picking up medicine boxes is unremarkable. It cannot backflip or dance. Its movements are slow, it has to plan a route before walking, and it stops to wait when someone crosses its path. It looks a bit clumsy. But this is often how industrial deployment begins: in an unglamorous setting, do one small thing that must be done right, do it ten thousand times without error, then talk about scale and other settings. Only when it can recognize, hold and place thousands of medicines on the shelf correctly is it worth talking about kitchens and living rooms. The medicine box is a gate it cannot go around. Anyone who wants into a home has to pass it first. There is no shortcut. **Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.** --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-pharmacy-robot-16-1105 Free to quote with attribution and a link to the original.