Learning and Planning with Structured Abstraction for Embodied Decision-Making

About the thesis

This thesis studies how structured abstractions can make learning and planning more efficient for embodied decision-making. It connects symmetry-aware representations, compositional world models, differentiable planning, and planning under perceptual uncertainty.

Research themes

• Structured representations and symmetry
• Compositional world modeling
• Differentiable and decision-time planning
• Acting under perceptual uncertainty