SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes
Researchers have developed SceneSmith, a novel agentic framework designed to generate realistic indoor environments for robot training simulations. This system uses a hierarchical approach with interacting VLM agents to create scenes from natural language prompts, incorporating asset generation and physical property estimation. SceneSmith produces significantly more objects than previous methods with minimal collisions and high physical stability, achieving high realism and prompt faithfulness in user studies. AI
IMPACT Enables more diverse and complex training environments for home robots, potentially accelerating their development and deployment.