Robotics
Spatial grounding for agents
that have to act in the physical world.
What we are actually solving
An agent planning a physical action needs to know whether the thing fits, whether it is reachable, and whether it stays put. Rowe answers those as an API call, over point clouds or over CAD, and returns typed output a planner can branch on.
How teams use it
01
Reachability and fit
Given an object and a target, get containment, clearance, and orientation as numbers a planner can threshold against.
task_type: fit
02
Stability estimation
Ask whether a placement holds: support area, contact, and the constraints that make it fail.
task_type: stability
03
Action and goal planning
Supply a goal state and get a structured plan grounded in the scene rather than in prose.
task_type: act
What you send us
- point_cloud_scene
- Images
- Sensor readings
- STEP
Files are parsed into a structured scene and reasoned over. Uploads are processed for the request, not retained as a library — persist the outputs on your side. Data retention
What we do not claim
Rowe is a reasoning service, not a controller. It runs off the robot, over the network, and carries no real-time or safety guarantees.
Try it on one of your own files.
The quickstart is a single curl call. If you would rather see it before you build against it, upload a part in the workbench.