Vision & Language for
Autonomous AI

VL4AI advances intelligent embodied AI for real-world robotic and autonomous systems. We study how multi-sensor and multimodal models can perceive, understand, reason, and act in complex dynamic environments, with a strong emphasis on interpretable, trustworthy, auditable, and safe embodied intelligence.

• Multimodal and multi-sensor perception, including 2D vision, 3D vision, sensor fusion, and embodied scene understanding.
• 3D reconstruction, semantic mapping, and world modelling for dynamic embodied environments.
• Compositional and neuro-symbolic reasoning for visual grounding, multimodal understanding, and decision making.
• Agentic embodied systems for planning, navigation, search, tracking, and mission-level autonomy.

Our goal is to build embodied AI systems that do more than perceive the world: they can reason over it, act within it, and support reliable real-world deployment through interpretable, trustworthy, and safe intelligence.