Trajectory Foundation Models

I develop universal trajectory foundation models that can learn from billion-scale worldwide traces and generalize across different cities and domains.

Key Works

  • UniTraj (NeurIPS 2025): A universal trajectory foundation model learned from billion-scale worldwide traces
  • OmniTraj (KDD 2025): Learning generalized trajectory models from omni-semantic supervision

Research Goals

  • Develop pre-trained models that can be fine-tuned for various downstream tasks
  • Explore cross-city transfer learning capabilities
  • Investigate efficient representation learning for large-scale trajectory data