(Ongoing work) KC-UKF is a multi-sensor fusion architecture that applies the Compatibility Field Estimation (CFE) framework to dynamically modulate measurement noise covariance in the Unscented Kalman Filter (UKF) update step. The CFE layer computes per-sensor health indicators via Normalized Innovation Squared (NIS) mapped through Gaussian kernels, then distributes trust across sensors using a softmax attention mechanism governed by an interaction matrix. This v0. 1. 0 release benchmarks KC-UKF against seven alternatives (Standard UKF, Adaptive Fading UKF, Sage-Husa UKF, Huber Robust UKF, IMM-UKF, VB-AQ-KC-UKF, and Wishart UKF) across three operationally distinct UAV scenarios and a maritime cross-domain validation, each evaluated over 100 Monte Carlo trials with Holm-Bonferroni corrected statistics. A multi-trajectory MCMC parameter-recovery study across all four scenarios establishes that KC-UKF's performance limitations on chronic-degradation and cross-domain regimes are parametric (fully recoverable via tuning), while the kinematic-model-mismatch limitation on ballistic re-entry is architectural (requiring process-model switching that CFE alone cannot provide). The deposit contains the main research report (KC-UKFResearchReport. pdf) including the CFE Core Algorithm as a self-contained appendix, and a source archive (kc-ukf-v0. 1. 0-source. zip) with the full LaTeX and Python codebase, raw Monte Carlo trial arrays (. npz), and a complete 7-tier evidence-based verification audit.
Kovid et al. (Sun,) studied this question.