PhySGA: A Physics-Space Grounded Agentic Framework for Generalizable and Interactive 3D Composition
PhySGA uses collaborating agents and physics simulation in Isaac Sim for complex 3D object assembly.
PhySGA is an agentic framework for interactive 3D object composition grounded in physical simulation rather than static language or diffusion priors. An Assembly Agent handles geometric planning and coarse placement, while a Critic Agent diagnoses simulator contact state and applies fine corrections. Action Routers execute physical primitives, and a self-evolving skill library stores trial-and-error experience. Experiments in Isaac Sim report gains in relation fidelity, physical plausibility, and visual quality for rigid, articulated, and deformable assemblies.
- Assembly and Critic agents plan steps and correct contact errors in closed loop.
- Action Routers turn plans into simulator primitives; a skill library reuses past trials.
- Supports rigid, articulated, and deformable objects, including hanging and hidden containment.
- Isaac Sim tests report better relation fidelity, physical plausibility, and visual quality.
From the original article
1 min read at arxiv.org3D object composition is pivotal to digital content creation and robot simulation, while also serving as a critical testbed for physical intelligence in foundation models. Previous methods either leverage large language models to directly predict object coordinates, or utilize generative diffusion priors to synthesize composite images for visual alignment, both of which rely on static priors in digital spaces.
By Weichen Lyu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.