Livgens is the I/O layer for Living Worlds: the tools, interfaces, integrations and applications of Living Worlds Lab. Real data is finite, and for anything new it never existed. Livgens represents anything real exactly: how a room carries sound, how a fabric falls, how a part takes a load, how light lands on a street. Living Worlds Lab gives it a lived lifecycle, one life or a billion, over a second or a century. Models train and test on what comes out, and every result is checked against the world.
Models have read most of what people ever wrote down. What is left is shrinking, locked away, or was never recorded at all: the rare failure, the building not yet built, the robot that has not yet moved. Training a model on its own output does not fill the gap. It wears away the rare cases first.
The public text that models learn from is expected to be used up between 2026 and 2032.
Epoch AIThe largest open robot dataset holds about a million episodes, pooled from 34 labs because no single lab could gather it.
Open X-EmbodimentA model trained on what models made loses the tails of the real world first, and does not get them back.
Nature, 2024Sources: Epoch AI, Open X-Embodiment, Shumailov et al., Nature.
To simulate something, you first have to say exactly what it is. That is Livgens: anything real, in any matter, held as the thing itself and not a picture of it. Living Worlds Lab takes it from there and runs it: how it behaves, how it changes, how its lives unfold, for one or for billions.
How a room or a venue carries it, seat by seat, as impulse responses you can listen to.
TwinWhat a camera sees from where it is mounted, with the truth behind every pixel.
RenderParts, structures and circuit boards from FreeCAD and KiCad, with their loads and checks.
RenderHow a fabric drapes and fits on a measured body, before it is cut.
RenderA room, a building or a city, from a drawing, photographs or a scan, carried as OpenUSD, IFC and CityJSON.
Render · SiteYour own measurements, and bodies fitted to any size.
Health · RenderHouseholds, customers and whole countries, living their lives over decades.
PeopleRobot and vehicle episodes, recorded as datasets a policy learns from, and the machine itself, measured by hardware nodes on the real thing.
School · Sense · PilotExact, with its limits.
Every thing carries its measurements and the range they hold for, so a simulation does not reach past what is known.
One, many or a billion.
Living Worlds gives it a lived lifecycle, over a second, a season or a whole life, as one life or a billion simple ones.
Data that did not exist.
What comes out is data a model can learn from and be tested against, in places where none was ever collected.
Held to the world.
Each result is registered before its data is read and scored once on measurements no model has seen. The misses are published too.
Simulation does not replace measurement. It makes a little of it go much further, and the results say exactly how far.
Power through feeders nobody had metered. With five real measurements, a model that learned from the simulation first made 19% less error than one trained on those five alone. At eighty, the two were level. A deliberately wrong simulation did worse than either.
met1,556 questions it had never seen, across thirty-one African countries: inside the survey's own noise in 30 of 31.
metMen's life expectancy forecast twenty years ahead and scored on 2025 figures nobody had read: within 1.09 years, with no lean either way.
metThe power through 40 unmetered feeders, every fifteen minutes for a year it had never seen: ahead of physics and a learned model in every season.
metA billion simple lives on one machine, and effects that only appear once there are enough of them.
metLivgens takes in the real world as well as describing it. Hardware nodes on a machine, a structure or a site take measurements. Each reading is verified when it arrives and kept with the node that took it. Those readings calibrate the twin and test it, and a plan that works in the twin can be staged and sent back out to the machine. Then the machine is measured again.
Livgens is for robotics and automation: from simulation and training in digital worlds, through to physical and digital interaction and control. Living Worlds is the digital worlds, where populations and individual agents are simulated, trained and taught. Livgens provides every integration and interface between those worlds and the physical one: measurements in; plans, commands and receipts out. Furcate is the substrate underneath.
Sensing and ingest are built. Control of machines is designed and staged, and is not yet switched on for anyone.
Hardware coverage today: Arduino Nicla Sense ME, ESP32-S3 and Raspberry Pi 5 are verified on hardware. NVIDIA Jetson Orin and Thor are built against vendor documentation and have not been run on a device. Each part, with its status
Machines already on a bus are read, never written: ROS 2, OPC UA, EtherCAT, MQTT, SocketCAN and Modbus are declared as observe-only integrations, built and tested against recorded traces and not yet run on a real device. Each integration, with its status
The machine and its surroundings run in a twin, through what could happen.
Render · SchoolA plan is trusted only inside the range the twin has been solved and checked for. Outside it, the twin says so and does not guess.
in the twinA new policy first runs in shadow: it proposes, the real controller acts, and the two are compared before it is given any say.
designedCommands go out under a signed policy that fixes what is allowed. Anything outside it is refused, and every command is kept in a record that cannot be quietly changed.
designedThe nodes report what actually happened. The twin is calibrated against it, and the loop starts again.
sensing builtMotion, orientation, magnetic field, pressure, temperature, humidity and air quality, plus electrical, light, flow and sound, and for machines position, velocity, torque, force, energy and status code. Each has one unit and a plausible range, and a reading outside it is refused, not clipped.
SenseTilt angle, vibration RMS, angular speed, field strength, dew point, pressure height change and mechanical power. Each is computed only in the regime where it is valid and refused outside it. A result carries the lowest level of its inputs.
SenseA system declares what it is for: compute, processing, physics, motion, measurement, environment, energy, fluid and thermal, or structure. It declares the quantities it measures, the actions it may take and the safe limits on each. An action is always bounded by those limits, and never outside what its twin has solved.
PilotA model is born into a robot, a machine, a piece of physical infrastructure or the world, and into other software, through the same interface. Everything starts in shadow mode: nothing is issued and a request to act is refused.
one interfaceTools sit inside the software you already use. Applications hold a whole place, a population or a body. Every one runs the change first, and asks before anything real happens.
Sound and music. The artist's collaborator in Ableton Live and on stage, with its own instruments.
early accessThe physical world in 3D. Built and simulated in Blender, Unreal Engine, FreeCAD, KiCad and RealityScan.
early accessLearning by living. Train an agent or a robot in measured worlds, with evidence it improves in the real one.
in developmentPlaces, machines and the hardware nodes on them, measured live in typed quantities with one unit each. Every reading verified and attributed. It reads and never writes back.
in developmentRobots, machines and vehicles. Simulation first, then staged control under a policy a person signs, always inside declared safe limits.
in developmentA place, from a room to a city or a river valley. Its buildings, networks, green space and water, kept live.
in developmentSoftware. A partner in your terminal and your browser that writes the change, runs your tests and shows its work.
early access
Simulation proposes. A person chooses. Nothing moves in the real world on a guess.
It proposes. You choose.
Each one shows what it would do and why, and waits for your word before it does it.
The real world asks a person.
A person sets and signs the policy a machine runs under, and a command outside it is refused. Anything new is shown to a person before it is sent.
Tried in a twin before it is tried for real.
Changes run in a simulation of the place or the machine first, and the result is shown with how sure it is.
Stricter where it matters most.
On power, water and roads, Livgens reads and advises. Changes go through the operator's own control systems and the people who run them.
The API and the SDK are the programmable layer under all of Livgens. Represent a thing, run it, and pull the data out from your own software, under the same rules: the same checks, and the same person asked before anything real happens.
Livgens's applications and tools are in early access. Tell us what you want to represent, and what you need it to do.