Tuning of evolutionary particle filtering approach for estimation of trajectory deviation

Abstract

An autonomous vehicle launched from moving platform is subjected to flexure-like disturbance and needs to stabilize itself before embarking on the target tracking phase. Control and guidance remains inactive during stabilization phase. Hence, estimating the deviation from the intended trajectory is important in the absence of control and guidance commands for trajectory correction. This estimation problem is here mapped to the misalignment estimation problem, with the intended trajectory data resembling mother data and the followed trajectory data resembling daughter data. The main requirement is fast convergence for early commencement of target tracking phase. Kalman filter and its variants fail to converge due to the inability to model time varying nonlinearities and flexure in the system model. Conventional particle filter converges, but computation time is high. This work explores evolutionary particle filter variants that may provide faster convergence through adaptive tuning of the range from where particle support points are selected. Such heuristics is based on the observed residual trend. Monte Carlo simulation results underline the importance of designing estimator based on adaptive tuned evolutionary particle filter algorithm.