PSTNet: Physically-Structured Turbulence Network

Dimensions overall score 8.0

• PSTNet achieves a mean miss-distance improvement of +2.8% with a 78% win rate and statistically significant effect size in guidance simulations.result 95%
  Evidencepartial

  PSTNet achieves a mean miss-distance improvement of +2.8% with a 78% win rate and a statistically significant effect size.

  Implicationpartial

  Directly stated in abstract with specific numeric results

  Verificationpartial

  partial

• PSTNet contains only 552 learnable parameters, requiring fewer than 2.5 kB of storage.technical 95%
  Evidencepartial

  The entire model contains only 552 learnable parameters, requiring fewer than 2.5 kB of storage

  Implicationpartial

  Directly stated in abstract with specific numeric values

  Verificationpartial

  partial

• PSTNet executes in under 12 seconds on a Cortex-M7 microcontroller.technical 95%
  Evidencepartial

  executing in under 12s on a Cortex-M7 microcontroller

  Implicationpartial

  Directly stated in abstract with specific performance metric

  Verificationpartial

  partial

• PSTNet embeds physics directly into its structure through four components: a zero-parameter backbone, regime-gated mixture of specialist sub-networks, Feature-wise Linear Modulation layers, and a Kolmogorov output layer.method 90%
  Evidencepartial

  PSTNet couples four components: (i) a zero-parameter backbone derived from Monin-Obukhov theory, (ii) a regime-gated mixture of specialist sub-networks supervised by Richardson-number-derived soft targets, (iii) Feature-wise Linear Modulation layers conditioning hidden representations on local air-density ratio, and (iv) a Kolmogorov output layer enforcing inertial-subrange scaling as an architectural constraint.

  Implicationpartial

  Directly described in abstract with specific architectural components

  Verificationpartial

  partial

• Classical spectral models encode climatological averages rather than instantaneous atmospheric state, and generic ML regressors offer adaptivity but provide no guarantee that predictions respect fundamental scaling laws.limitation 85%
  Evidencepartial

  Classical spectral models encode climatological averages rather than the instantaneous atmospheric state, and generic ML regressors offer adaptivity but provide no guarantee that predictions respect fundamental scaling laws.

  Implicationpartial

  Directly stated in abstract as motivation for the work

  Verificationpartial

  partial

• PSTNet is validated on 340 paired six-degree-of-freedom guidance simulations spanning three vehicle classes and six operational categories with real-time satellite weather ingestion.method 90%
  Evidencepartial

  We validate PSTNet on 340 paired six-degree-of-freedom guidance simulations spanning three vehicle classes (Mach 2.8, 4.5, and 8.0) and six operational categories with real-time satellite weather ingestion.

  Implicationpartial

  Directly stated in abstract with specific validation details

  Verificationpartial

  partial

• Encoding domain physics as architectural priors yields a more efficient and interpretable path to turbulence estimation accuracy than scaling model capacity.result 80%
  Evidencepartial

  Our results demonstrate that encoding domain physics as architectural priors yields a more efficient and interpretable path to turbulence estimation accuracy than scaling model capacity

  Implicationpartial

  Directly stated conclusion in abstract, though comparative claim requires some inference

  Verificationpartial

  partial

• PSTNet establishes a viable drop-in replacement for legacy look-up tables in resource-constrained, safety-critical on-board guidance systems.market 85%
  Evidencepartial

  establishing PSTNet as a viable drop-in replacement for legacy look-up tables in resource-constrained, safety-critical on-board guidance systems.

  Implicationpartial

  Directly stated conclusion in abstract, though 'viable' implies some judgment

  Verificationpartial

  partial

Competitive landscape

PSTNet is a lightweight, physically-structured neural network for real-time atmospheric turbulence estimation, offering a drop-in replacement for legacy systems in aircraft guidance.