PhySGA: A Physics-Space Grounded Agentic Framework for Generalizable and Interactive 3D Composition
PhySGA uses simulation feedback to correct 3D assemblies as they are built, tackling hanging, settling under gravity and hidden containment.
An Assembly Agent plans construction steps and coarse adjustments, while a Critic Agent diagnoses simulated contact dynamics and guides finer corrections. Action Routers translate those decisions into executable simulator operations for rigid, articulated and deformable objects. A reusable skill library captures lessons from earlier trial and error to improve success rates and reduce exploration attempts. The authors report better relation fidelity, physical plausibility and visual quality than existing methods in Isaac Sim experiments.
ArXiv · AI/CL/LG's note
An Assembly Agent plans construction steps and coarse adjustments, while a Critic Agent diagnoses simulated contact dynamics and guides finer corrections. Action Routers translate those decisions into executable simulator operations for rigid, articulated and deformable objects. A reusable skill library captures lessons from earlier trial and error to improve success rates and reduce exploration attempts. The authors report better relation fidelity, physical plausibility and visual quality than existing methods in Isaac Sim experiments.
ArXiv · AI/CL/LG's note
score 5