Dynamic Worlds, Dynamic Humans: Generating Virtual Human-Scene Interaction Motion in Dynamic Scenes.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Real-world scenes are dynamic, but current human-scene interaction models assume static environments.
• This static assumption limits the realism and applicability of generated human motions.

Purpose of the Study:

• To introduce Dyn-HSI, the first cognitive architecture for dynamic human-scene interaction.
• To enable virtual humans to perceive, remember, and interact with changing environments.

Main Methods:

• Dyn-HSI integrates three components: Dynamic Scene-Aware Navigation (vision), Hierarchical Experience Memory (memory), and Human-Scene Interaction Diffusion Model (control).
• A new dynamic benchmark, Dyn-Scenes, was created by extending existing datasets.

Main Results:

• Dyn-HSI consistently outperforms existing methods in generating high-fidelity human-scene interaction motions.
• The architecture effectively handles dynamic scene changes and improves motion quality and generalization.

Conclusions:

• Dyn-HSI represents a significant advancement in generating realistic human-scene interactions within dynamic environments.
• The proposed cognitive architecture offers a more grounded approach for virtual human behavior modeling.