How it works.
What can be trained here, what each kind of run is for, how a twin stays in step with the real thing, and which tools it works with.
What lives here
Living systems and built systems, in one world.
A river and its fish, a chair on a test line, a street at evening, a piece of music: each lives on its own clock, and each leaves a record.

Rivers and reefs
Fish, seagrass and corals with a sensor net, seen through the water.

Interiors and furniture
A room, its furniture and its design twin, tried before it is built.

Art and design
Paintings, sculpture and generative forms, growing in a studio.

Sound and music
Instruments and rooms, with the waves they make drawn as they travel.

Products and lines
A product, its line and its twin, from the first sample to the shelf.

Tests and structures
A bridge, a wing, a chair: loaded, shaken and weathered on purpose.
What is trained
What learns by living here.
Each lives a life in a world, and each leaves with what that life taught it.
World models
Generate scenes with world models: a warehouse at night, a road in rain. A plausibility check scores generated motion against gravity, ground contact and energy.
Agents
Give an agent a whole life and it remembers what a person told it. Four years on, it predicts their answers better than a persona does, by 2.5%.
Populations
Let a million lives meet over thirty years. Effects appear that no single life shows.
Robots
A robot arm learns to grasp across thousands of parallel attempts.
Data
Datasets made from lived runs: trajectories, sensor scenes and generated video, each with the record of how it was made.
Simulation and training
Every kind, with an example.
One engine, many ways to use it.
Policy training
Reinforcement learning for a manipulator, a quadruped or a vehicle: train, play back, roll out, benchmark and export the policy.
Imitation and fine-tuning
A humanoid policy fine-tuned on recorded demonstrations, in the formats robot teams already use.
Synthetic data
Scenes and sensor views of rare situations, generated and rendered, kept with how they were made.
Populations
One million lives for ten years, replayable by anyone.
Digital twins
A twin, kept in step with the real thing.
A twin mirrors a real machine, facility or site. Readings from the real one keep it calibrated.

Mirror
A real machine or site, with its parts and its limits, running beside the real one.
Calibrate
Readings are admitted inside the range the twin was solved for. Outside it, the twin says so.
Learning
What it learns, and where it goes.
Learning is what comes out of a world. It leaves as something you can use.
01
What is learned
Policies and skills. Behaviours. Forecasts with a stated range.
02
What it can tell you
Causal effects are estimated and tested by refutation, the parameters of a simulator are inferred from its own runs, and predictions carry conformal intervals.
03
How it leaves
Policies export for the real machine. Forecasts run beside live readings. Datasets travel with their record.
04
Where it is applied
A robot cell, a plant, a grid, a service. Ipnops connects it.

Formats
Open formats in and out.
The worlds, the lives and the records are Living Worlds' own. Scenes, tasks and policies move in and out through open formats, and what comes back is kept with its record.
Scenes and assets
Scenes move in and out as standard scene description files, and simulation-ready assets are imported into them.
Robot learning
Tasks to train, play back, roll out, benchmark and export policies.
Physics
GPU physics for rigid bodies, cloth and deformable media.
World models
World models that generate and transfer scenes from a prompt or a source video.
Humanoid policies
Humanoid policy inference and fine-tuning on robot datasets.
Rendering
Camera rendering of simulated scenes.
Hardware, runtimes and the connection to real machines are on ipnops.ai.

Which world will you live in?
What grows here is meant to reach the real world. Start with a world of your own.