Reimagining service delivery in the agentic era with Google Public Sector
cloud.google.com·19h ago
AI Research
arXiv CS.AI·Jun 10·5 min read
Mobility Anomaly Generation using LLM-Driven Behavior with Kinematic Constraints
TL;DR
A significant gap exists in the availability of datasets containing annotated human mobility anomalies, which are crucial for spatial data mining. To address this, a novel generative framework was developed that uses Large Language Model (LLM) agents to create realistic trajectory anomalies while adhering to physical constraints.
✦ Why It Matters
Engineers and researchers can leverage this framework to generate realistic mobility anomaly datasets for better analysis and modeling.
Key Takeaways
- Utilize LLMs to create realistic mobility anomalies.
- Incorporate kinematic constraints for spatial validity.
- Enhance datasets with context-aware spatial noise models.
- Address the scarcity of annotated trajectory anomaly datasets.
How It Works
The framework generates mobility anomalies by modifying baseline simulated trajectories using LLM agents. These agents introduce behavioral anomalies, which are then adjusted for spatial validity through map-constrained routing.
The integration of a spatial noise model further enhances the realism of the generated data by simulating variations in GPS sensor accuracy based on environmental factors.