Others teach robots the skill. We teach them the place.
A robot that can pick, inspect or drive still has to learn the site it works in. We build that site as a simulation, keep it current and let you test before the robot arrives.
- Scene v3 loaded · OpenUSD
- Crop stage set · 5 of 6
- Row clearance · aisle 2 · pass
- Canopy contact · row 3 · flagged
A skill is general. A site is specific.
A robot needs both. Skills data is available from robot makers and skills data companies. Data about your own site has to be built for that site.
What to do
- Grasping, walking, folding, opening a door
- Learned from demonstrations, teleoperation and motion capture
- General: the same skill applies in many places
- Supplied by robot makers and skills-data companies
Where to do it
- This building, this greenhouse, this plant yard
- Built from CAD, BIM, lidar and drone surveys of the real place
- Specific: every site is different and keeps changing
- Supplied by Kaiinos, and kept current
A robot that has learned to pick, open a door or read a gauge still has to find the right shelf, fit through the right doorway and reach the right valve on your site. That is the part we build.
If a robot works there, we can simulate it.
Every site has its own layout and its own changes. These are some of the environments we can build as simulation scenes.
Racking moves, pallets land in aisles, stock levels change
AMRs, mobile manipulators, humanoids
Doors open and close, furniture moves, people pass through
Humanoids, service robots, telepresence robots
Hot-work zones open and close, equipment is added or moved
Quadrupeds, field robots, drones
Panels get dirty or damaged, vegetation grows, weather changes
Drones, cleaning robots, UGVs
Tides, depth, vessel traffic and quay layouts
Autonomous vessels, drones
Crops grow through their stages, rows and terrain vary
Field robots, UGVs, drones
Your real site, ready to simulate.
We build and maintain the scene. Your team, or your skills-data partner, brings the robot and its skills.
Site scenes built from real data
We build the simulation scene from your CAD, BIM (IFC), lidar point clouds and drone surveys, with assets labelled so a robot knows what each object is. Scenes are delivered in OpenUSD for NVIDIA Isaac Sim and can be exported to Gazebo.
source · IFC model + lidar scan assets labelled · rows, rails, doors → greenhouse.usd ready for Isaac Sim
Conditions that change over time
Real sites do not stand still. Crops grow, stock moves and zones close. We model those states so you can test against them, for example representative crop growth stages with plant height, canopy width and leaf area for each stage.
crop stage 5 of 6 canopy width and height updated → row clearance recalculated
Test scenarios before the robot arrives
Run missions against the scene and see what fails: routes that do not fit, rows that are too narrow, contact with crops or equipment, blocked aisles and closed zones. Problems show up on a screen rather than on site.
scenario · aisle 2 at stage 5 clearance pass · 1 canopy contact → result stored against scene v3
Kept current with the digital twin
When the scene is linked to a Kaiinos digital twin, confirmed changes on site become a new scene version. Your tests keep matching the place the robot will actually work in, not last year's drawings.
layout change confirmed · bay 4 scene v3 → v4 → scenarios re-run on v4
Synthetic data, as an extension
For teams that train perception models, rendered images with labels can be produced from the same scene. This is scoped separately from the base scene.
render pass · rows 1-4 labels · fruit, stem, leaf, rail → dataset exported for training
Humanoids, robots and autonomous systems.
Humanoids
Rehearse picking, carrying and moving through rooms, doors and corridors built for people.
Humanoids →Robotics
Test inspection rounds, robot cells and AMR routes against the site as it is today.
Robotics →Autonomous systems
Plan drone, vehicle and vessel missions over terrain, water and infrastructure before launch.
Autonomous systems →From site data to first test run.
Start with one site and one mission. Prove it there, then scale.
Pick a site and a mission
We agree one site, one robot and the task that matters most.
Build the scene
We turn your site data into a labelled simulation scene, or capture the site first.
Add conditions and scenarios
Growth stages, layouts and edge cases are set up for testing.
Run, review, keep current
Your team runs the tests, and the scene is updated as the site changes.
Formats your team already uses.
Scenes run on your own simulation set-up. We are also building a hosted option, so teams can run tests on demand.
Frequently asked questions
Do you train the robot's skills?
No. Skills such as grasping, walking or folding come from your own team or from companies that supply skills data through demonstrations, teleoperation and motion capture. Kaiinos supplies the site those skills are used in, so you can test them against the real layout, assets and conditions.
Which simulators do you work with?
Scenes are built in OpenUSD for NVIDIA Isaac Sim and can be exported to Gazebo. If your team uses another tool, we agree the export format at the start of the project.
How do you model crops and other things that change?
For greenhouses, a crop growth model converts biomass into plant height, canopy width and leaf area for each growth stage. The scene uses representative stages and does not predict daily growth. The same approach works for other site states, such as stock levels, closed zones or seasonal terrain.
Can we run tests without our own GPU workstations?
Today scenes are delivered to run on your own simulation set-up. We are building a hosted option so teams can run tests on demand. Tell us if that matters for your project and we will include it in the pilot discussion.
How close is the simulation to the real site?
The scene is built from your site data, such as CAD, BIM, lidar and drone surveys, so layout, dimensions and asset positions follow those sources. We agree the level of detail and what is in and out of scope before work starts.
Does the scene stay up to date?
Yes, when it is linked to a Kaiinos digital twin. Confirmed changes on site become a new scene version, so your tests keep matching the place the robot will actually work in.
Bring us the skill. We'll bring the place.
Tell us which site and which robot you want to test.
Start a pilothello@kaiinos.ai