V3I11P28

Adaptive and Robust Filtering for Systems with Random Observation Delays: Pseudo-Measurement Estimation with Delay Compensation for Acoustic Navigation

Vu Hoai Nam*1

Abstract

Acoustic measurements are the main source of position information for autonomous underwater vehicles (AUVs). Sound travels at about 1500 m/s, so every acoustic observation arrives after a random delay that grows with range and becomes comparable to the sampling step at distances of kilometres. Ignoring these delays biases the estimate, optimal Bayesian filtering is computationally prohibitive, and in practice the prior information about the initial position, the vehicle velocity and the noise distributions is incomplete. This article turns the framework of Bosov and Uryupin for stochastic systems with random observation delays into a concrete and computationally simple filter based on linear pseudo-measurements. The delays are compensated by an augmented state that stores lagged positions; the initial position and the mean velocity are obtained by weighted least squares; the velocity is identified online; and the measurement noise scale is adapted from the innovations after a short transient. A Monte Carlo study with 100 runs of 600 s, three surface observers and bearing and elevation accuracy of 0.2° shows that in the steady state delay compensation reduces the RMS position error from 8.44 m to 3.23 m (from 7.98 m to 1.71 m with four times better angular accuracy). Between 60 and 120 s the least-squares start gives 6.84 m, against 25.34 m for a broad prior without velocity information. The error with uniform, Laplace and Student-t noise of the same variance differs from the Gaussian case by less than the Monte Carlo uncertainty. When the measurement noise triples, the adaptive filter estimates the noise scale factor as 9.0 (true value 9) and reduces the error over 400–600 s from 7.02 m to 6.29 m, close to the worst-case design (6.15 m).

Keywords:

autonomous underwater vehicles; acoustic navigation; random observation delays; pseudo-measurements; Kalman filtering; adaptive filtering; robust filtering; Monte Carlo simulation