When an image becomes an equation: calculating the sensitivity of a spatial design without recourse to the test bench

The protocol is structured into three progressive phases (in silico, minimal physical validation, mission scale) with explicit and quantified GO/NO-GO/PIVOT criteria, which limits the risks of post hoc interpretation and favours a rational decision at each stage.

The idea of treating a Monte Carlo differential rendering pipeline as a first-class executable physical model within an MBSE flow is conceptually appealing and fills a genuine gap: the MBSE literature (e.g. [2024] Bibliometric Analysis of MBSE) does not propose a mechanism for propagating gradients through differentiable physical models, and the differential rendering literature (e.g. [2020] Differentiable Rendering: A Survey) does not position itself within a systems engineering context. The proposed chaining (architectural parameters → scene parameters → rendering → requirement loss → gradients) constitutes an original methodological contribution.

The idea of using automatic differentiation through a Monte Carlo rendering engine to obtain gradients of requirements with respect to architectural parameters is conceptually appealing and goes beyond finite-difference or surrogate-model approaches.

Niche but high-value market: space agencies (ESA, NASA, CNES) and optical payload integrators (Airbus Defence and Space, Thales Alenia Space, OHB) spend tens of millions of euros per year on thermal and optical test campaigns to validate pointing and image-quality requirements. A tool capable of computing loss gradients at requirement level with respect to architectural parameters (FOV, sunshield, integration time) would drastically reduce the number of bench iterations, with a potential ROI of 5 to 10x on a programme such as PLATO or ARIEL.

The hypothesis combines two domains in high demand (differentiability of rendering pipelines and MBSE for space systems), which creates an attractive interdisciplinary narrative for European calls with a digital and space component.