Human spaces were never designed for robots.
Humanoids are walking into warehouses, factories, hospitals, public buildings and the yards between them, all built around people: narrow aisles, doors, stairs, kerbs, shelves at arm height, and people everywhere. To be useful, they need to understand those spaces the way we do.
- Tote 14 located · rack B3, shelf 2
- Grasp point sent to robot
- Route clear of 2 people
To be useful, a humanoid has to understand the site.
Precise maps indoors and out, down to doors and handles
Centimetres matter when a robot walks through a doorway, steps off a kerb or reaches for a handle. A floor plan drawn years ago, or a scan with no labels, is not enough.
Knowing what objects are and where they belong
A robot that sees a box still has to know which box it is, where it should be, and what to do when it is somewhere else.
Awareness of people and how spaces are used over the day
Corridors, lifts, entrances and yards fill and empty on a daily rhythm. Robots that know it can work around people instead of through them.
A site your humanoid can read.
Kaiinos supplies the model of the site and the scenes to test in. Your robot's own software keeps control of planning and motion.
Semantic digital twins of whole sites, indoors and out
Every building, floor and room, plus the yards, paths and ramps between them, as one measured model: walls, doors, shelves, stairs, lifts, kerbs and slopes, each labelled as what it is. The robot's software can ask for a room, a door or a loading bay instead of working it out from raw geometry.
clear width at Door D-14 → 0.92 m · robot body 0.68 m · fits
Object and shelf-level labels for picking and manipulation
Shelves, slots and objects carry a position and the place they belong, plus how they are handled: which way a door opens, lever or knob, which drawer pulls out. Picking and tidying tasks get a reference to work against.
where is tote 14? expected: rack B3, shelf 2 → grasp point sent to robot
Occupancy and flow patterns learned from sensor history
Site sensor history shows which corridors, lifts, entrances and crossings are busy, and when. Tasks can be planned into quieter hours and onto quieter routes.
corridor C2 · weekdays busiest 08:30-09:15 and 12:00-13:00 → schedule restocking 10:00-11:30
Simulation environments to train and test tasks before go-live
The same site is exported as an OpenUSD scene for NVIDIA Isaac Sim, or for Gazebo, so tasks are tested in the real layout before the robot arrives on site.
Warehouse A · all floors → OpenUSD objects and shelves labelled → scene ready for Isaac Sim
Built for the machines entering human spaces.
General-purpose humanoids
Two-armed robots taking on varied tasks in spaces built for people.
Mobile manipulators
Wheeled bases with an arm, picking and placing across a site.
Service robots
Delivery, cleaning and guidance in hospitals, offices and public buildings.
Telepresence robots
Remote presence that has to move safely through busy rooms.
From first scan to first deployment.
Start with one site and one machine. Prove it there, then scale.
Pick a site and a machine
We agree on one environment, one robot and the mission that matters most.
Capture and model
We capture the site and build its digital environment on the Kaiinos platform.
Simulate the missions
Routes, tasks and edge cases are tested before the robot arrives.
Deploy and keep it current
The robot goes live on a living digital twin that updates as the environment changes.
Pick your site.
Picking and restocking from labelled shelves
Shelf, bay and slot labels with object home locations give a humanoid or mobile manipulator a reference for picking, putting away and spotting gaps. Aisle traffic patterns from sensor history help schedule work around forklifts and pickers.
Line-side delivery and material handling
Kitting areas, line-side racks, doors and step changes between halls are modelled, so material-handling tasks can be tested in simulation first. Keep-out zones around machinery and work cells are kept as editable layers.
Fetch and carry between wards
Corridors, lifts, door types and restricted areas are mapped floor by floor. Flow patterns show when corridors near wards, theatres and entrances are busiest, so delivery tasks can be planned for quieter times.
Service robots in offices and halls
Reception areas, meeting rooms and public halls change use through the day. A semantic twin with people-flow layers helps operators of service and telepresence robots plan tasks that avoid crowding and respect restricted zones.
Moving goods between buildings
Loading docks, yards, ramps, kerbs and the paths between halls are modelled with the same labels as the spaces indoors, so a robot carrying goods from one building to the next works from one model. Satellite and aerial data cover the wider site, using the same geospatial work behind our mapping projects.
Frequently asked questions
Does Kaiinos control or navigate the humanoid robot?
No. Path planning and motion stay with your robot's own software, or your integrator's. Kaiinos gives that software a better picture of the site: what each part is, where objects belong, where the robot can and cannot go, and when spaces are busy. Two things we are developing add live perception on top. Kaiinos Node, our edge sensing device, detects changes on the site and gives robots fixed points to localise against where satellite positioning fails. A robot perception engine will take in camera and lidar feeds from your robots and keep the shared twin current.
What is a semantic map for robots?
A semantic map adds meaning to geometry. Instead of only walls and free space, it labels doors, handles, shelves, stairs, lifts and objects, and records where items belong and how they are used. A humanoid can then be given a task such as fetching from a specific shelf, rather than simply reaching a coordinate.
What do you need from us to build a site twin?
Whatever site data you have: CAD drawings, BIM models in IFC, site plans, earlier scans and floor plans. If sources are missing or out of date, we capture the site with lidar. For people-flow layers we need access to site sensor history, such as occupancy or people-counting data, for the areas the robots will work in.
Which simulators can use the twin?
We export scenes in OpenUSD for NVIDIA Isaac Sim, and scenes can also be exported for Gazebo. They carry the site geometry and semantic labels, so your team can test manipulation and task sequences in a model of the real site before a pilot starts.
Does the twin update itself when the site changes?
Today the twin is updated on a regular schedule, with change history so you can compare versions. Live updates from robot cameras and lidar are part of the robot perception engine, which we are building towards. It will sync twins to robots, detect local changes and update the shared twin in near real time.
From our field notes
Bring us the machine. We'll bring the world it works in.
Tell us which site and which tasks you are piloting.
Start a pilothello@kaiinos.ai