Swarm robotics
Also known as Robot swarms, Collective robotics, Multi-robot swarms
Swarm robotics coordinates many simple robots through local sensing and communication rather than a central controller. Laboratory swarms have reached 1,024 robots, while the largest drone shows, military drone programmes and warehouse fleets mostly rely on central control.
In plain language
A swarm robot is deliberately simple. It can sense only a few nearby robots and makes small mistakes, so no single unit can do much on its own. The idea is that a large group, following simple local rules, can still build a useful outcome, in the same way that a flock of birds or a colony of ants has no leader directing every move.
The best-known research example is the Harvard Kilobot. In 2014 a team in Radhika Nagpal's lab at Harvard used 1,024 small robots to form two-dimensional shapes, including a five-pointed star and the letter K. Other uses are more common today but work differently. Drone light shows fly many drones in the sky at once, but a single computer usually plans and controls the whole display. Warehouses run large fleets of mobile robots that carry goods, managed by central software. These are useful, but they show how hard it is to keep many machines working together safely.
Technical explanation
Swarm robotics studies how large numbers of robots with limited sensing and communication can coordinate through local interaction. Because there is no single controller, the group behaviour has to emerge from rules run on each unit. Designers usually work with three constraints: each robot senses only nearby neighbours, communication is short range, and behavioural rules are simple. The aim is robustness, so that losing a unit does not stop the task, and scalability, so that adding units does not require new central planning.
The 2014 Harvard study shows what this looks like in practice. The robots were myopic and had no overhead view of the group. The team reported that errors such as traffic jams and off-course robots became more frequent as the group grew, and that nearby robots detected these problems and cooperated to correct them. Individual units were less reliable: they had trouble driving in a straight line and their distance sensing varied from robot to robot. The team also said that simulation alone was not enough, and that testing on physical robots was needed to find failures that appear only at scale.
Other field results are smaller in scale. A 2016 study reported that neural controllers evolved in simulation for up to ten aquatic surface robots transferred to a real, uncontrolled water environment and performed comparably to simulation on homing, dispersion, clustering and monitoring tasks. Large drone displays are a different engineering problem. The Guinness World Records category for drones airborne simultaneously counts drones flown from a single computer outdoors, and the 2026 record stood at 22,580 drones. That is scale through central control rather than decentralised swarming, and it is the main reason this profile treats drone shows as a neighbouring technology rather than a proof of swarm autonomy.
What the evidence says
Each statement is typed by the strength of its evidence and linked to the source that supports it.
Demonstrated Shown in a published experiment, test or shipping product.
In 2014 Harvard researchers reported 1,024 Kilobots forming two-dimensional shapes, including a five-pointed star and the letter K, with commands sent to all robots at once using infrared light.
Sources: [1] · Release textEach Kilobot was myopic: it could see only a few neighbours and had no overhead view of the whole group.
Sources: [1] · Release textThe Kilobot team reported that errors such as traffic jams and off-course robots became more common in large groups, that nearby robots detected and corrected some of them, and that individual robots were less reliable, with trouble driving straight and variable distance sensing.
Sources: [1] · Release text, limitations sectionEHang's 22,580-drone flight over Hefei on 3 February 2026 is the entry Guinness World Records lists for the most multirotors or drones airborne simultaneously from a single computer, outdoors.
Sources: [3] · Record detailsThe Guinness record title specifies drones airborne from a single computer. The record page does not describe the control method, so the category reads as central control of the flight rather than decentralised swarming.
Sources: [3] · Record titleIn a real, uncontrolled water environment, up to ten aquatic surface robots ran neural controllers evolved in simulation, and the authors report performance comparable to simulation on homing, dispersion, clustering and monitoring tasks.
Sources: [7] · Abstract · medium confidenceReported, unverified Stated by a company or team but not independently confirmed.
In August 2023 the US Department of Defense set a goal for its Replicator initiative to field thousands of attritable autonomous systems within roughly 18 to 24 months. In September 2025, current and former officials suggested hundreds had been delivered, and a former Replicator-1 director said thousands more were on contract. Defense officials did not verify these remarks.
Sources: [4] · Article body · medium confidenceAmazon says it has deployed its one millionth robot, across more than 300 facilities worldwide.
Sources: [6] · Company announcement · medium confidenceAlmost 90 of the 1,000 drones in a Sydney Vivid show fell in May 2026; the operator blamed radio interference. The Australian Transport Safety Bureau has opened an investigation, but the reporting gives no findings.
Sources: [5] · Article body · medium confidenceLimitations and open problems
- Individual robots are unreliable, so the group must tolerate failures. The 2014 team reported that robots had trouble moving in a straight line and that distance sensing varied from robot to robot.
- Errors become more common as groups grow. The 2014 study found that traffic jams and off-course robots increased with group size, and local correction did not remove them.
- Local sensing and communication limit what a swarm can know. Robots with no overhead view depend on rules that encode the global goal, which makes the design harder to check.
- Most field evidence is small. The aquatic robot study used up to ten robots, while the largest decentralised swarm reported here, 1,024 robots, was tested in a laboratory.
- Safety is a major constraint for flying swarms. In May 2026 almost 90 of the 1,000 drones in a Sydney show fell, and the operator blamed radio interference. An investigation has been opened, but no findings were reported.
- Public evidence on military swarms is thin. The Replicator programme's fielded numbers, cost and status are not independently verified, and officials declined to share details.
Possible future uses
Ideas, not predictions. None of these has been demonstrated.
- Search and monitoring of disaster areas by many low-cost robots that keep working when individual units are lost.
- Environmental monitoring of lakes or coastal waters by small groups of surface robots, building on the aquatic robot trial.
- Collective construction of simple structures by robots that build together without a central plan, extending the 2014 shape-forming work.
Timeline
- 14 Aug 2014Harvard researchers report a swarm of 1,024 Kilobots that form two-dimensional shapes, including a five-pointed star and the letter K. Sources: [1]
- Mar 2016PLoS ONE publishes a real-world trial of evolved controllers on up to ten aquatic surface robots. Sources: [7]
- Aug 2023The US Department of Defense sets a goal for the Replicator initiative to field thousands of attritable autonomous systems within about two years. Sources: [4]
- 26 Sept 2024A Damoda show over Shenzhen sets the Guinness record for drones airborne simultaneously from a single computer outdoors, with 10,197 drones. Sources: [2]
- 3 Feb 2026EHang flies 22,580 drones over Hefei, the record Guinness World Records lists in this category. Sources: [3]
- May 2026Almost 90 of 1,000 drones fall during a Sydney Vivid show; the operator blames radio interference. Sources: [5]
Futures this could help bring about
Glossary
- Swarm robotics
- The study of many simple robots that coordinate through local sensing and communication, without a central controller.
- Decentralised control
- A control approach in which each unit makes decisions from its own sensing and nearby contacts, rather than from one central plan.
- Attritable system
- A machine designed to be cheap enough that losing or destroying it in use is acceptable, so it can be replaced rather than protected.
- Kilobot
- A small, low-cost research robot from Harvard, used in large swarm experiments.
Sources
- university press release
A self-organizing thousand-robot swarm
Harvard School of Engineering and Applied Sciences, 14 Aug 2014. Accessed 21 Oct 2026. - news
A dazzling display: China's record-breaking drone spectacle with over 10,000 drones
Guinness World Records, 17 Oct 2024. Accessed 21 Oct 2026. - other
Most multirotor/drones airborne simultaneously from a single computer (outdoors)
Guinness World Records, 5 Feb 2026. Accessed 21 Oct 2026. - news
DOD touts 'successful transition' for Replicator initiative, but questions linger
Brandi Vincent. DefenseScoop, 3 Sept 2025. Accessed 21 Oct 2026. - news
How did almost 90 drones crash during Sydney's Vivid festival?
ABC News (Australia), 27 May 2026. Accessed 21 Oct 2026. - company announcement
Amazon deploys over 1 million robots and launches new AI foundation model
Amazon, 1 Jul 2025. Accessed 21 Oct 2026. - preprint
Evolution of Collective Behaviors for a Real Swarm of Aquatic Surface Robots
Miguel Duarte, Vasco Costa, Jorge Gomes et al.. arXiv (journal version: PLoS ONE 11(3), 2016, e0151834), Mar 2016. DOI: 10.1371/journal.pone.0151834. Accessed 21 Oct 2026.
Revision history
3 changes since this profile was created.