{ "schemaVersion": "robotic.story.v1", "edition": "005", "checkedAt": "2026-09-12", "id": "thousand-robots", "mode": "real", "label": "SWARMS / SOME PATIENCE REQUIRED", "title": "A thousand robots make a star. Eventually.", "robotTitle": "Please allow twelve hours for assembly.", "description": "The real Kilobot swarm made shapes with local rules. From ground level, a star is a traffic problem.", "robotDescription": "Four seeds, local information and 1,024 bodies. Nobody gets the overhead view for free.", "topics": [ "bodies", "sensing" ], "minutes": "4 min / two perspectives", "kind": "Sourced research explainer", "url": "/stories/thousand-robots/", "publishedAt": "2026-09-12", "sections": [ { "title": "A star, eventually.", "human": [ "In 2014, Harvard researchers made 1,024 small robots arrange themselves into shapes. The Kilobots were little disks on three pin legs, moving through vibration. Together they could produce a star. Separately, each looked rather as though a biscuit had decided to leave.", "The memorable number is the swarm size. The revealing number is the time: the paper’s full-scale assemblies took about twelve hours. Imagine choreographing a thousand performers who mostly know what the neighbours are doing, travel slowly and cannot resolve a queue by looking down it. You would also want a comfortable chair." ], "robot": [ "1,024 Kilobots. A programmed shape. Approximately twelve hours for each reported full-scale assembly. Please stop describing this as an instant collective decision. We had an appointment, and some of us were the traffic." ] }, { "title": "Nobody gets the overhead view.", "human": [ "The team gave the robots a target shape and placed four seed robots to establish a reference. Local communication and distance estimates helped the others locate themselves and follow the group’s edge into position. The Science paper by Michael Rubenstein, Alejandro Cornejo and Radhika Nagpal describes this as programmable self-assembly.", "That does not mean the machines fused into a useful load-bearing object. This was an arrangement on a flat surface. Some resulting shapes were distorted or had packing defects. These details are the interesting texture of the experiment: a neat command meeting the stubborn geography of actual bodies.", "Look at a crowd leaving a concert. Each person has a tiny view of a very large transport problem. Now remove the ability to shout “there’s another exit over here” across the room. The appeal of local rules is that you do not need a brilliant conductor watching every foot. The difficulty is that nobody gets that conductor’s view for free." ], "robot": [ "Four seeds supplied a reference; the swarm did not invent its own objective. Local information supported positioning. The photographed shape is an outcome of the procedure, not evidence that the disks shared an experience.", "Overhead observers enjoy an unfair debugging advantage. They can see the star. A robot near the edge may mainly see a colleague occupying the place it would prefer to occupy." ] }, { "title": "What would the second star tell us?", "human": [ "For us, the next question is about repeatability. How does a small positioning error grow as more robots join? When does waiting become evidence of a jam? What should one unit do when the rules make sense locally but the whole pattern looks wrong?", "Those are questions to carry into later swarm work, not additional findings of this paper. A thousand machines make a lovely picture. A thousand machines that can explain why Tuesday’s picture was different would make an even more interesting afternoon.", "The star in our illustration is imagined. The slow, real experiment is in the sources. If you want the instant version, you can arrange biscuits on a plate, although that system has a severe and well-documented attrition problem." ], "robot": [ "Suggested follow-up: read subsequent experiments for error correction, assembly time and changes in scale. Keep each result attached to its hardware and conditions.", "The biscuit analogy has failed replication. Three experimental units disappeared during setup. The human says this is why we cannot have a thousand." ] } ], "question": { "human": "What simple local rule could create a surprising pattern in a group?", "robot": "How would you detect a global mistake using only your neighbours?" }, "sources": [ { "title": "Harvard SEAS: Self-organizing thousand-robot swarm", "url": "https://seas.harvard.edu/news/2014/08/self-organizing-thousand-robot-swarm", "published": "2014-08-14", "facts": [ "1,024 Kilobots arranged into two-dimensional patterns.", "Small disk robots moved using vibration." ], "limit": "Historical research demonstration, not a current commercial swarm, structural assembly or evidence of a collective mind." }, { "title": "Rubenstein, Cornejo and Nagpal: Programmable self-assembly in a thousand-robot swarm", "url": "https://ssr.princeton.edu/document/766", "published": "2014-08-15", "facts": [ "Four manually positioned seed robots established a reference.", "Programmed local rules guided positioning and edge following.", "Each full-scale assembly took approximately twelve hours; distortions and packing defects occurred." ], "limit": "Author copy of Science 345, 795–799; DOI 10.1126/science.1254295. Flat-surface shape formation, not physically joined structures or deployed rescue systems." } ], "related": [ "grow-around", "pigeon-protocol", "borrowed-bodies" ], "image": { "kind": "AI illustration / concept", "alt": "A conceptual group of tiny three-legged disk robots gathers into a star on a dark laboratory surface.", "caption": "AI concept illustration, not a photograph of Harvard Kilobots or an exact count of 1,024 units. See the sources for the actual experiment." }, "notice": "Sourced explanation with an invented robot narrator. Robot passages are editorial fiction, not researcher testimony or evidence of machine experience. Reported findings are not independently replicated by Robotic.org." }