TeLoGraF: Temporal Logic Planning via Graph-encoded Flow Matching
Yue Meng, Chuchu Fan
2025-05-05
Summary
This paper talks about TeLoGraF, a new AI system that helps plan actions over time by understanding and following complicated rules, using advanced graph-based techniques.
What's the problem?
Planning tasks that have to follow specific rules over time, like making sure certain things happen in a certain order, is really hard for AI, especially when the environment is complex or changes a lot.
What's the solution?
The researchers built a system that uses graph neural networks and a method called flow-matching to help the AI figure out the best way to meet all the rules quickly and accurately, and they showed that it works better and faster than older methods.
Why it matters?
This matters because it makes AI much better at handling real-world planning problems, like robotics, scheduling, or controlling systems, where following the right steps at the right time is crucial.
Abstract
TeLoGraF uses Graph Neural Networks and flow-matching to solve general STL specifications in various simulation environments, outperforming baselines in STL satisfaction rate and inference speed.