Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics
About
While humans effortlessly discern intrinsic dynamics and adapt to new scenarios, modern AI systems often struggle. Current methods for visual grounding of dynamics either use pure neural-network-based simulators (black box), which may violate physical laws, or traditional physical simulators (white box), which rely on expert-defined equations that may not fully capture actual dynamics. We propose the Neural Material Adaptor (NeuMA), which integrates existing physical laws with learned corrections, facilitating accurate learning of actual dynamics while maintaining the generalizability and interpretability of physical priors. Additionally, we propose Particle-GS, a particle-driven 3D Gaussian Splatting variant that bridges simulation and observed images, allowing back-propagate image gradients to optimize the simulator. Comprehensive experiments on various dynamics in terms of grounded particle accuracy, dynamic rendering quality, and generalization ability demonstrate that NeuMA can accurately capture intrinsic dynamics.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Future state prediction | Spring-GS (test) | CD0.16 | 28 | |
| Object Dynamics Grounding | Synthetic Object Dynamics Grounding Dataset | BouncyBall Score3.34 | 6 | |
| Initial state generalization | Gorilla Real-world | PSNR29.78 | 5 | |
| Initial state generalization | Real-world dataset Chick2 | PSNR28.92 | 5 | |
| Initial state generalization | Real-world dataset Chick1 | PSNR30.73 | 5 | |
| Initial state generalization | Real-world dataset Mandarin | PSNR31.85 | 5 | |
| Initial state generalization | Real-world dataset Peanut | PSNR30.7 | 5 | |
| Initial state generalization | Real-world dataset Rabbit | PSNR28.3 | 5 | |
| Initial state generalization | Real-world dataset RBball | PSNR30.74 | 5 | |
| Intrinsic Dynamics Consistency | SandFish synthetic (test) | Chamfer Distance1.07 | 4 |