Marginalized particle filter for spacecraft attitude estimation from vector measurements

Abstract

An algorithm based on the marginalized particle filters (MPF) is given in details in this paper to solve the spacecraft attitude estimation problem: attitude and gyro bias estimation using the biased gyro and vector observations. In this algorithm, by marginalizing out the state appearing linearly in the spacecraft model, the Kalman filter is associated with each particle in order to reduce the size of the state space and computational burden. The distribution of attitude vector is approximated by a set of particles and estimated using particle filter, while the estimation of gyro bias is obtained for each one of the attitude particles by applying the Kalman filter. The efficiency of this modified MPF estimator is verified through numerical simulation of a fully actuated rigid body. For comparison, unscented Kalman filter (UKF) is also used to gauge the performance of MPF. The results presented in this paper clearly demonstrate that the MPF is superior to UKF in coping with the nonlinear model.

Steady-State Marginalized Particle Filter for Attitude Estimation

“…From this result, we can verify that the steady-state MPF performs better than the standard PF, but not necessarily be better than the EKF. The second observation is consistent with the results reported in [9,12]. The bias estimate errors are plotted in Figure 5, which shows comparable performance from each filter.…”

Attitude Estimation of a Quadcopter with one fully damaged rotor using on-board MARG Sensors

“…A method similar to particle filters is very computationally heavy (Crassidis et al (2007b)), although it provides a very high accuracy of results. Cheng and Crassidis (2004), Liu et al (2007) and Carmi and Oshman (2009) have developed particle-filter based attitude estimation techniques, but as stated above, they are computationally costly. Furthermore, Bonnabel et al (2009b) suggested a new filter called Invariant Extended Kalman Filter (IEKF).…”

Rao-Blackwellized particle filter with vector observations for satellite three-axis attitude estimation and control in a simulated testbed

“…The procedure reduces the variance of Monte Carlo estimates and is applicable when, conditioned on a set of states, the remaining ones are linear and Gaussian (Doucet, 1998). Liu et al (2007) have investigated a similar approach known as the marginalized PF applied to attitude and rate-gyro bias estimation with vector observations and also resorting to rate-gyro measurements. Here, the state vector has been partitioned into two groups: one with attitude-related components and the other with angular rate components.…”

“…Samples have been taken from the second group with nonlinear dynamics, whereas the components in the first group, which are also nonlinear, are estimated using an Extended Kalman filter. Therefore, unlike the model studied at Liu et al (2007), the system model here is not conditionally linear, but the Rao-Blackwellization approach becomes applicable by use of some mild approximations. Moreover, a significant reduction of the number of particles to attain an estimation accuracy much similar to that of the standard particle filter has been attained by concatenating pseudo-measurements of the angular rate to the measurement vector.…”