Jump to: overview, details, publications
Embodied AI
Vision-Language Graph Models for Embodied Navigation
Integrating vision-language embeddings with global semantic graphs to enable adaptive navigation in unseen environments.
Overview
Integrating vision-language embeddings with global semantic graphs to enable adaptive navigation in unseen environments.
Details
This initiative explores a hybrid framework that combines vision-language representations with structured semantic graphs to enable agents to reason, navigate, and act in complex environments.
Current work
- Unified multimodal world models that integrate perception, memory, and planning.
- Scalable graph-based reasoning for open-world navigation.
- Robust transfer from simulation to real-world deployment.