2026/07/01 by Nicolas Merlinge, Karim Dahia
Computer Science · #Target Tracking and Data Fusion in Sensor Networks #Distributed Sensor Networks and Detection Algorithms #Distributed Control Multi-Agent Systems
paper · doi:10.1016/j.ijar.2026.109783
Bayesian estimation with non-injective measurement models and unknown measurement error bounds is a tricky problem. When the measurement model is too complex to be embedded in an algorithm, the estimation filter may not model all the physical disturbances of the observation and the measurement error becomes uncontrolled (both in terms of distribution and support). This paper introduces an adaptive version of the classic Box Regularized Particle Filter (BRPF) by deriving the Adaptive Approximate Bayesian Computation technique (A2BC) to the bounded error framework. The approach consists of an optimized constraint relaxation for the box contraction step in terms of Efficient Sampling Size (ESS). The consistency of the proposed approach is theoretically evaluated by deriving closed-forms of the estimation discrepancy and of the ESS criteria. A2BC-BRPF is proved to provide a more consistent posterior distribution than a non-adaptive BRPF algorithm. Numerical results are presented on a severely nonlinear aerial terrain navigation scenario and the A2BC-BRPF method is shown to outperform state of the art algorithms both in terms of accuracy and robustness to unknown measurement errors.