One Robot
One indoor mobile platform is the baseline, rather than multiple hardware families at once.
/manual/field
Field Manual
The first version is intentionally narrow. This manual page records the intended job, the operating assumptions behind it, and the limits that keep the design coherent.
Mission Profile
- Map a room or apartment.
- Navigate to a named destination within the known indoor map.
- Identify a supported object near the target area.
- Pick it up with a mobile arm.
- Move it to a low-height drop box and place it down safely.
This is a bounded indoor workflow, not an open-world household robot promise.
First-Version Focus
One Software Spine
A ROS 2-based architecture is the reference path for integration, launch layout, and service ownership.
One Local Workflow
Data capture, review, upload, and later model updates are designed around a user-owned local host.
Operating Assumptions
| Assumption | Meaning in practice |
|---|---|
| Bounded indoor navigation | The robot is expected to work inside mapped rooms, not as a general outdoor rover. |
| Small supported object set | Perception and grasping are aimed at a deliberately limited class of household objects. |
| Local network ownership | Training data review and model update paths should work inside an operator-controlled environment. |
| Reproducible documentation | Key assumptions should be written down before the implementation surface expands. |
Out Of Scope For Now
- Arbitrary object handling with no scope limits.
- Cloud-dependent operation as the default deployment path.
- Unrestricted open-domain voice interaction as a baseline interface.
- End-to-end reinforcement learning as the primary control architecture.
- Broad performance claims before stable hardware and software baselines exist.
New Object Workflow
The project also reserves space for a local recording path. When the robot encounters an unsupported object, an operator should be able to capture data, move it to a local host, and reuse it later for dataset growth and model updates.
This is a local extension workflow, not a promise of autonomous open-set learning during normal robot operation.
Documentation Priority
At the current stage, the documentation itself is part of the system definition. Interfaces, subsystem boundaries, issue routing, and safety language are being written down before the robot behavior is treated as reproducible.