MAMHOI: Factorizing Scene-Aware Human-Object Interaction through Affordances
MAMHOI separates scene understanding from motion generation for realistic human-object interactions. It uses a scene model to predict where an interaction is feasible and a motion model to synthesize the actual movement, linked by an affordance interface. This factorization allows learning from different data types without needing paired human-object-scene data, resulting in more physically plausible and realistic scene-aware interactions.