GAT Platform
GAT Documentation
GAT is a cloud-edge tri-model framework that fuses learning, analytical constraints and logical control for verifiable robotic task planning.
Overview
This page is the documentation shell for the GAT World site. The final library will be populated from Markdown files supplied by the backend, while this route already uses a real documentation layout: persistent section navigation, article typography, source-oriented examples and an on-page table of contents.
Tri-model loop
GAT documentation should explain the system as a task-completion framework. The model loop combines learned scene understanding, analytical feasibility checks and logical control rules before execution is sent to the robot runtime.
| Layer | Responsibility | Output |
|---|---|---|
| Learning model | Interprets scene state, task intent and candidate operations. | Task sequence proposal |
| Analytical model | Checks spatial, temporal and physical constraints. | Validated plan state |
| Logic controller | Chooses execution order, replanning boundaries and failure handling. | Executable task policy |
Cloud-edge runtime
The docs should separate cloud-side planning from edge-side execution. Cloud services can maintain task context, run heavier reasoning and aggregate simulation evidence; the edge runtime should preserve collision constraints, local sensing and low-latency robot control.
- Cloud planning: mission decomposition, simulation rollouts and global consistency checks.
- Edge execution: local perception, collision-safe actuation and task-state feedback.
- Closed-loop update: execution evidence returns to the planner for refinement.
Markdown source
The eventual documentation library can be driven by front matter and article bodies. The page is ready for a generated navigation index, article routes and Markdown rendering.
---
title: Tri-model Framework
section: GAT Platform
order: 2
---
# Tri-model Framework
Explain the learning model, analytical model and logic controller as one task-completion pipeline.Next steps
Once the backend provides Markdown, the placeholder sections should become real article routes with search, versioned navigation and source-backed diagrams. The current page keeps the documentation surface clean until that content lands.