Research map / Video / Video generation
DyPSI: Dynamic Physics Sensing via Joint Field and Sensor-Trajectory Generation
Yizhou Zhang, Panqi Chen, Lei Cheng, Ting Zhang, Jianlong Li, Jianfeng Shi, Shikai Fang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes DyPSI, a method for dynamic physics sensing that jointly generates dense spatiotemporal fields and sensor trajectories from sparse observations.
Problem
Reconstructing dense spatiotemporal physical fields from sparse sensor observations in evolving environments, while adapting where future sensors should be placed; existing methods assume static sensing layouts or optimize placements independently per frame.
Approach
Jointly generates the physical field and sensor trajectories over time instead of using a fixed layout or independent per-frame sensor placement, so that sensing adapts dynamically to the evolving field.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| generate sensor trajectories dynamically and jointly with fields instead of static optimized placement | DyPSI vs PhySense-opt · Turbulent Flow, 50 sensors | relative L2 error: 0.1558 -> 0.1352 | Improved |
| generate sensor trajectories dynamically instead of static placement | DyPSI vs PhySense-opt · Car aerodynamics, 15 sensors | relative L2 error: 0.0386 -> 0.0311 | Improved |
| generate sensor trajectories dynamically instead of static placement | DyPSI vs PhySense-opt · Sea Temperature, 15 sensors | relative L2 error: 0.0439 -> 0.0402 | Improved |
| use Matern-3/2 GP-correlated noise instead of i.i.d. noise (ablation) | DyPSI · Turbulent Flow, 50 sensors | relative L2 error: 0.1515 -> 0.1352 | Improved |
| use Matern-3/2 kernel instead of RBF kernel (ablation) | DyPSI · Turbulent Flow, 50 sensors | relative L2 error: 0.1489 -> 0.1352 | Improved |
| use joint generation of fields and sensors instead of separate diffusion models (ablation) | DyPSI · Turbulent Flow, 50 sensors | relative L2 error: 0.1421 -> 0.1352 | Improved |
| swap teacher reconstructor from PhySense to Senseiver or DiffusionPDE for trajectory supervision (ablation) | DyPSI-opt · Turbulent Flow, 30 sensors | relative L2 error: 0.4570 -> 0.4527 (Senseiver); 0.7904 -> 0.7116 (DiffusionPDE) | Improved |
| use grid or min-max sampling instead of random sensor placement (ablation) | conventional placements · Turbulent Flow | relative L2 error | No clear effect |
| use per-frame oracle placement without kinematic constraints (ablation) | sensor placement · Turbulent Flow, 50 sensors | velocity violation rate: 8% -> 89% | Hurt |
| vary SPF bandwidth sigma within a factor of two (ablation) | DyPSI · reconstruction error | relative L2 error | No clear effect |
Closest papers by meaning
- Flow Field Reconstruction with Sensor Placement Policy Learning (NeurIPS 2025 · similarity 0.87)
- PhySense: Sensor Placement Optimization for Accurate Physics Sensing (NeurIPS 2025 · similarity 0.86)
- Generating Full-field Evolution of Physical Dynamics from Irregular Sparse Observations (NeurIPS 2025 · similarity 0.86)
- StreamPhy: Streaming Inference of High-Dimensional Physical Dynamics via State Space Models (NeurIPS 2026 · similarity 0.85)
- Christoffel-DPS: Optimal sensor placement in diffusion posterior sampling for arbitrary distributions (NeurIPS 2026 · similarity 0.85)
- Adaptive Flow Matching for Resolving Small-Scale Physics (ICML 2025 · similarity 0.85)
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.
