GAT World is a research and engineering effort for verifiable robotic task completion. We are building a bridge from the Generative Adversarial Tri-model framework into cloud simulation, analytical constraint validation and edge robot execution.

The work is not a generic autonomous-driving demo. It is a system for decomposing tasks, checking physical and logical constraints, and keeping execution grounded in the scene state returned by the robot.

Research is better when grounded

Paper-to-system translation matters. We care about reading the source papers correctly, preserving the tri-model mechanism and turning the method into task-level behavior that can be inspected, tested and improved.

Robots need constraints

Execution must respect the room. The analytical model should make spatial, temporal and collision constraints explicit before the robot moves. A task is complete only when the execution state satisfies the plan, not when a visual trajectory looks plausible.

Cloud and edge should co-design

Reasoning and control have different latency budgets. Cloud services can support planning, simulation and documentation; edge systems should handle perception feedback, local actuation and safety-critical task updates.

Join us

We are preparing collaboration tracks for the paper team, robotics engineers, simulation builders and documentation contributors. Final team profiles, open roles and contact channels will be added when the project materials are ready.