Professor, University of Pennsylvania
2 papers at NeurIPS 2025
Contextualizing MLLM-based agents with grounded scene graphs boosts their performance.