TeLoGraF: Temporal Logic Planning via Graph-encoded Flow Matching
Yue Meng, Chuchu Fan·May 01, 2025
Summary
TeLoGraF, utilizing Graph Neural Networks and flow-matching, excels in complex task-solving with STL specifications, outperforming baselines in satisfaction rate. It adeptly handles intricate STLs, showing robustness to out-of-distribution specifications, and addresses limitations of existing methods. Key contributions include STL planning, domain applications in robotics and formal logic, and integration with machine learning for enhanced performance. Recent advancements focus on learning from demonstrations, reactive synthesis, robust control, and multi-agent motion planning. TeLoGraF demonstrates its capabilities in the Franka Panda environment with blue trajectories for single and multi-goal tasks.
Introduction
Background
Overview of Graph Neural Networks (GNNs) and their applications
Introduction to Signal Temporal Logic (STL) and its significance in task specification
Objective
Highlighting TeLoGraF's performance in solving complex tasks with STL specifications
Discussing how TeLoGraF outperforms baselines in satisfaction rate
Method
Data Collection
Gathering STL specifications for various tasks
Collecting data on task environments and scenarios
Data Preprocessing
Transforming STL specifications into a format compatible with GNNs
Preprocessing task environments for efficient model training
Key Contributions
STL Planning
Describing the process of planning tasks based on STL specifications
Discussing the integration of GNNs and flow-matching in STL planning
Domain Applications
Exploring applications in robotics, focusing on task execution and motion planning
Highlighting the use of TeLoGraF in formal logic for automated reasoning
Machine Learning Integration
Discussing how TeLoGraF leverages machine learning for enhanced performance
Exploring the role of learning algorithms in improving task satisfaction
Recent Advancements
Learning from Demonstrations
Exploring how TeLoGraF learns from expert demonstrations
Discussing the benefits of this approach in task learning and generalization
Reactive Synthesis
Describing the process of synthesizing reactive systems from STL specifications
Highlighting TeLoGraF's role in generating efficient and adaptive systems
Robust Control
Discussing the application of TeLoGraF in designing robust control strategies
Exploring how it handles uncertainties and disturbances in task execution
Multi-Agent Motion Planning
Exploring TeLoGraF's capabilities in coordinating multiple agents for complex tasks
Discussing the integration of STL specifications in multi-agent systems
Case Study: Franka Panda Environment
Single and Multi-Goal Tasks
Describing the Franka Panda environment setup
Presenting TeLoGraF's performance in single and multi-goal tasks
Blue Trajectories
Analyzing the significance of blue trajectories in visualizing task execution
Discussing how they represent successful task completion and satisfaction of STL specifications
Basic info
papers
formal languages and automata theory
robotics
machine learning
artificial intelligence
Advanced features