Optimal Control under Quantised Measurements - A Particle Filter and Reduced Horizon Approach
(2017)Department of Automatic Control
- Abstract
- This thesis covers the optimal control of stochastic systems with coarsely quantised measurements. A particle filter approach is used both for the estimation and control problem. Three main families of particle filters are examined for state estimation, standard SIR filters, SIR filters with generalised sampling and auxiliary filters. A couple of different proposal distributions and weight functions were examined for the generalised SIR and auxiliary filter respectively. The choice of proposal distribution had the greatest impact on performance but the unrivalled best filter was achieved with a combination of generalised sampling and the auxiliary particle filter. For the problem of control the particle filter was used for cost-to-go... (More)
- This thesis covers the optimal control of stochastic systems with coarsely quantised measurements. A particle filter approach is used both for the estimation and control problem. Three main families of particle filters are examined for state estimation, standard SIR filters, SIR filters with generalised sampling and auxiliary filters. A couple of different proposal distributions and weight functions were examined for the generalised SIR and auxiliary filter respectively. The choice of proposal distribution had the greatest impact on performance but the unrivalled best filter was achieved with a combination of generalised sampling and the auxiliary particle filter. For the problem of control the particle filter was used for cost-to-go evaluation by forward simulation in time. Simplifications of the full dynamic programming problem were done by reducing the time horizon resulting in M-measurement feedback policies and a new M-measurement cost feedback policy. One-measurement feedback and M-measurement cost feedback was examined for M 4 and although probing behaviour was observed none of the examined controllers managed to outperform a certainty equivalent controller. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8915326
- author
- Vikström Morin, Martin
- supervisor
-
- Anton Cervin LU
- Bo Bernhardsson LU
- organization
- year
- 2017
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6031
- ISSN
- 0280-5316
- language
- English
- id
- 8915326
- date added to LUP
- 2017-08-18 11:10:35
- date last changed
- 2017-08-18 11:10:35
@misc{8915326,
abstract = {{This thesis covers the optimal control of stochastic systems with coarsely quantised measurements. A particle filter approach is used both for the estimation and control problem. Three main families of particle filters are examined for state estimation, standard SIR filters, SIR filters with generalised sampling and auxiliary filters. A couple of different proposal distributions and weight functions were examined for the generalised SIR and auxiliary filter respectively. The choice of proposal distribution had the greatest impact on performance but the unrivalled best filter was achieved with a combination of generalised sampling and the auxiliary particle filter. For the problem of control the particle filter was used for cost-to-go evaluation by forward simulation in time. Simplifications of the full dynamic programming problem were done by reducing the time horizon resulting in M-measurement feedback policies and a new M-measurement cost feedback policy. One-measurement feedback and M-measurement cost feedback was examined for M 4 and although probing behaviour was observed none of the examined controllers managed to outperform a certainty equivalent controller.}},
author = {{Vikström Morin, Martin}},
issn = {{0280-5316}},
language = {{eng}},
note = {{Student Paper}},
title = {{Optimal Control under Quantised Measurements - A Particle Filter and Reduced Horizon Approach}},
year = {{2017}},
}