--- title: "FloatForm turns robot boats into the building blocks of the structure" locale: "en" url: "https://irz.fr/en/articles/floatform-robot-boats-en" markdown_url: "https://irz.fr/en/articles/floatform-robot-boats-en.md" category: "tech" tags: ["robotics", "swarm", "infrastructure", "water", "MIT"] published_at: "2026-08-19T08:21:00.000Z" author: "Léa Perrin" translation: "https://irz.fr/fr/articles/floatform-robot-boats-fr.md" --- # FloatForm turns robot boats into the building blocks of the structure Eight 21 cm boats assemble and travel as one structure. FloatForm moves most coordination from a global planner to neighbouring robots. A temporary floating bridge does not necessarily have to travel as a bridge. FloatForm starts from that useful premise: distribute material, motors and part of the computation across a fleet of small square boats, and assemble the structure only when it is needed. The MIT system uses eight modules, each 21 cm square. Every unit carries four miniature thrusters, sensors, a controller and a magnetic latching mechanism.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709)[2](https://www.nature.com/articles/s41467-026-74527-6) In the test pool they start scattered, converge on a target geometry, connect, detach, reconfigure and then travel as one larger assembly. It is easy to jump straight to temporary docks, stages, emergency platforms or bridges, although the research becomes more useful once those future renderings are set aside for a moment. FloatForm starts from a more practical question: **how can many modules coordinate while avoiding one central computer that has to calculate every movement for every robot?** ## Eight boats, not eight remote-controlled toys Each module is an autonomous vehicle measuring 0.21 × 0.21 m, with a hull 0.14 m high.[2](https://www.nature.com/articles/s41467-026-74527-6) Four miniature thrusters arranged in an X give the boat omnidirectional motion, covering forward and sideways translation as well as rotation in place. At that scale, the thrusters generate large forces compared with the boat's low inertia, so early prototypes were twitchy and prone to abrupt rotation at low speed. The team answered with fins under the hull, increasing hydrodynamic drag and giving the controller a calmer physical system to work with.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709)[2](https://www.nature.com/articles/s41467-026-74527-6) This detail breaks the fantasy of a system solved mainly in software: part of the stability comes from passive geometry under the robot, because no control algorithm gets to negotiate the basic behaviour of water away. ## The latch has to tolerate a bad approach Docking is already a precision problem on a table; on water, two modules trying to connect disturb each other through their own thrusters and the waves they create. FloatForm puts an origami-inspired auxetic mechanism inside each hull, where one central servo moves permanent magnets on all four sides at once.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709)[2](https://www.nature.com/articles/s41467-026-74527-6) The arrangement can capture a neighbouring module across gaps of roughly 10 to 15 cm, giving the boats some tolerance for an imperfect approach.[2](https://www.nature.com/articles/s41467-026-74527-6) Once connected, a 3D-printed gearbox holds the latch mechanically instead of continuously powering an electromagnet.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709) The battery pays for connection and release, not for every hour the modules remain attached; in a floating structure, that distinction is likely to matter quickly. ## The central planner does not vanish, it gets thinner Many modular robot systems rely on a global planner that knows the target shape, assigns positions and calculates collision-free trajectories. This works with small fleets, but computation and communication grow heavier as more modules wait for globally scheduled turns, so FloatForm uses a hybrid architecture instead.[2](https://www.nature.com/articles/s41467-026-74527-6) A central component still handles occasional high-level tasks, including final position assignment to refine the geometry, while routine navigation, collision avoidance and convergence towards the target are mostly handled onboard through exchanges with immediate neighbours.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709)[2](https://www.nature.com/articles/s41467-026-74527-6) The result is that robots can move in parallel instead of following one global sequence. The scalability claim rests on that redistribution of work: local computation depends mainly on a robot's neighbourhood rather than the total fleet size, while the physical experiments use eight modules and the simulations extend the framework to 64.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709)[2](https://www.nature.com/articles/s41467-026-74527-6) ## Promising results, not magic MIT reports that four-robot missions completed autonomously in 90% of ten trials, while the figure fell to 70% with eight robots.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709) Depending on the configuration, assembly took roughly four to eight minutes. Those numbers are more useful than the phrase "programmable infrastructure" because they show where the system actually is. It works often enough to demonstrate the approach, while a larger fleet already exposes more failure modes. The team also built recovery behaviours: a robot that temporarily loses its position can rejoin the structure, and modules caught in a deadlock can shake themselves free before trying again.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709) For a swarm, robustness means absorbing those local failures while the rest of the structure keeps working. ## A test pool still leaves a long way to a canal The scale-up gap remains substantial. The lab robots use ultrasonic positioning with four fixed beacons around a 4.0 × 2.7 × 1.2 m pool.[2](https://www.nature.com/articles/s41467-026-74527-6) A real canal would need a different localisation stack, such as GPS or vision. The latches and hulls would also have to withstand much larger forces. To be sure, the researchers acknowledge a hard limit: small boats cannot handle highly disturbed water, and vehicle size remains tied to the magnitude of waves and currents it can tolerate.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709) Scaling means revisiting latching, sensing, propulsion, stiffness and safety together rather than simply enlarging the same robot. The urban bridge in the concept art does not exist yet. What the lab has demonstrated comes earlier in the chain: autonomous modules can find, join and recover from some local failures while coordination remains largely distributed. ## Structure becomes a state rather than an object A conventional platform is manufactured, transported and used in roughly the same form. FloatForm proposes another sequence in which generic units travel separately, assemble for one function and return to the pool afterwards. Its potential value comes as much from autonomy as from **reversible form**. The same inventory of modules could theoretically become a platform, a dock, a barrier or a dispersed sensing fleet. At small scale, MIT has already demonstrated assembly, separation, reconfiguration and collective motion.[1](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709)[2](https://www.nature.com/articles/s41467-026-74527-6) There is a lot of water between eight squares in a test pool and urban infrastructure. Even so, the question is worth keeping: if the building blocks can already move, find one another and latch, perhaps the final structure only needs to exist at the moment it is used. ## References 1. [MIT News, Tiny robot boats build floating structures](https://news.mit.edu/2026/tiny-robot-boats-build-floating-structures-0709) 2. [Nature Communications, Self-reconfiguring modular robotic boats](https://www.nature.com/articles/s41467-026-74527-6)