The study distinguishes physical-system degradation from observation degradation across complementary NASA and ESA telemetry sources. The central claim is intentionally narrow: explicit observation reliability may improve calibration and robustness of spacecraft health inference under degraded multivariate telemetry.
y_t = h(x_t) + ε_t, ε_t ~ p(ε | r_t)
where x_t denotes latent physical system state, y_t the observed telemetry, h(·) the measurement process, ε_t observation error, and r_t an explicit representation of observation reliability. The key point is that abnormal telemetry can arise from the spacecraft, the observation process, or both.
p(x_t | y₁:t, r₁:t) ∝ p(y_t | x_t, r_t) · p(x_t | y₁:t−1, r₁:t−1)
The posterior health estimate conditions on both measurements and their reliability context. This is the public scientific idea under test; it does not expose the private Decision-RUL formulation or any held assurance mathematics.
Δcal = ECE_base − ECE_reliability
where ECE is expected calibration error. A positive Δcal under degraded telemetry would indicate that explicit reliability information improves probability calibration relative to a baseline that sees telemetry alone.
Δrobust = M_reliability,degraded − M_base,degraded
where M is a pre-declared evaluation metric appropriate to the experiment, such as discrimination, detection or probabilistic scoring. The NATRO research investigates whether reliability-aware inference degrades more gracefully when observations become incomplete, corrupted or ambiguous.
NASA ADAPTCONTROLLED EXPERIMENTPhysical fault insertion and observation-corruption testbed for identifying whether reliability-aware methods can separate sensing problems from system problems.
SMAP / MSL TELEMANOMREPRODUCIBILITYHistorical labelled anomaly benchmark retained for comparison with established anomaly-detection work.
ICESAT-2 ATL02 V7CURRENT ORBITAL TELEMETRYLarge heterogeneous HDF5 engineering telemetry with timing, instrument, spacecraft and quality context.
MARS 2020 MEDAEXTREME ENVIRONMENTMultisensor environmental and engineering observations with native temporal and quality context from Perseverance operations on Mars.
ESA-ADB / ESA-ADEXTERNAL VALIDATIONReal operational satellite telemetry with curated anomalies, rare-event context and mission splits for independent validation.
National Aeronautics and Space Administration — source institution for ADAPT, ICESat-2, Mars 2020 MEDA and the historical SMAP/MSL benchmark lineage.
OFFICIAL SOURCE →NASA Ames Research CenterADAPT controlled aerospace electrical-power testbed and fault-injection experiments.
OFFICIAL SOURCE →NSIDC DAAC / NASA EarthdataDistribution and metadata infrastructure for ICESat-2 ATL02 Version 7.
OFFICIAL SOURCE →NASA Planetary Data SystemArchive infrastructure for Mars 2020 MEDA calibrated, derived and engineering products.
OFFICIAL SOURCE →European Space AgencyOperational spacecraft anomaly benchmark lineage used for independent external validation.
OFFICIAL SOURCE →Research and engineering contributions across spacecraft telemetry, aerospace and energy systems, machine learning, physical intelligence, explainability and large-scale data infrastructure.