Research map / Probabilistic and scientific modeling / Dynamical systems
Graph Networks as Learnable Physics Engines for Inference and Control
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use a graph network instead of an MLP forward model | Forward dynamics prediction · Six of seven simulated control systems | prediction error | Helped |
| Use a graph network instead of an MLP forward model | Forward dynamics prediction · Pendulum | prediction error | No clear effect |
| Train the GN on Swimmer6 using random and DDPG trajectories | GN-based MPC · Swimmer6 | distance to target after 700 steps | Helped |
| Use a two-GN forward model instead of a single IN/GN (ablation) | GN forward model · Forward prediction | prediction performance | Helped |
| Use a recurrent forward model instead of a feed-forward model (ablation) | GN forward model · Real JACO | 100-step rollout error | Helped |
| Provide an identifiable 20-step phase with control inputs (ablation) | GN system identification model · Pendulum with variable hidden lengths | 100-step rollout error | Helped |
| Remove control inputs during the system-identification phase (ablation) | GN system identification model · Pendulum with variable hidden lengths | 100-step rollout error | Hurt |
| Provide a system-identification module instead of no identification information (ablation) | GN forward model · Parametrized systems | 100-step rollout error | Helped |
| Give the model true static parameters instead of inferred parameters (ablation) | GN forward model · Parametrized systems | 100-step rollout error | Helped |
| Use an ID phase from a different system instance than the prediction phase (ablation) | GN system identification model · Parametrized systems | 100-step rollout error | Hurt |
| Include Cheetah in mixed-system training (ablation) | Multi-system GN forward model · Multiple systems | prediction error | Hurt |
| Use SVG(1) instead of model-free SVG(0) (ablation) | SVG agent · Swimmer6 | performance | Helped |
Closest papers by meaning
- A Graph Dynamics Prior for Relational Inference (AAAI 2024 · similarity 0.88)
- Graph Mamba Operator: A Latent Simulator for Interacting Particle Systems (NeurIPS 2026 · similarity 0.88)
- Benchmarking Structural Inference Methods for Interacting Dynamical Systems with Synthetic Data (NeurIPS 2024 · similarity 0.87)
- SEGNO: Generalizing Equivariant Graph Neural Networks with Physical Inductive Biases (ICLR 2024 · similarity 0.87)
- Graph Neural Networks Gone Hogwild (ICLR 2025 · similarity 0.87)
- PGODE: Towards High-quality System Dynamics Modeling (ICML 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.