Difference between revisions of "EECI08: Packet-Based Estimation and Control"

From Murray Wiki
Jump to navigationJump to search
m
Line 4: Line 4:
 
This lecture describes how to extend results in estimation and control to the case where the information between sensing, actuation and computation flows across a network with possible packet loss and time delay.  We begin with the estimation problem, summarizing the results on Sinopoli et al on Kalman filtering with intermittent data, which uses average convergence as a stability metric.  An alternative formulation is to use almost sure convergence, which gives improved results for lossy networks.  Finally, we extend the results on estimation to the control setting, summarizing approaches in the cases where receipt of packets are acknowledge (TCP-like) or not acknowledged (UDP-like).
 
This lecture describes how to extend results in estimation and control to the case where the information between sensing, actuation and computation flows across a network with possible packet loss and time delay.  We begin with the estimation problem, summarizing the results on Sinopoli et al on Kalman filtering with intermittent data, which uses average convergence as a stability metric.  An alternative formulation is to use almost sure convergence, which gives improved results for lossy networks.  Finally, we extend the results on estimation to the control setting, summarizing approaches in the cases where receipt of packets are acknowledge (TCP-like) or not acknowledged (UDP-like).
  
{| width=100% border=1
+
==  Lecture Materials ==
|- valign=top
 
| width=50% |
 
==== Outline ====
 
<ol type="A">
 
<li>Kalman filtering with intermittent observations</li>
 
* Problem motivation and setup
 
* Mathematical preliminaries (Jensen's inequality)
 
* Main results: upper and lower bounds
 
<li>Probabalistic state estimation with packet drops</li>
 
* Probabalistic bounds versus expected value (example)
 
* Performance versus data loss tradeoff
 
<li>Packet-based control</li>
 
* TCP-like networks
 
* UDP-like networks
 
</ol>
 
| width=50% |
 
 
 
====  Lecture Materials ====
 
 
* Lecture slides: {{eeci-sp08 pdf|L7_pbc.pdf|Packet-Based Estimation and Control}}
 
* Lecture slides: {{eeci-sp08 pdf|L7_pbc.pdf|Packet-Based Estimation and Control}}
 
* Lecture notes: {{ncsbook|fusion|Chapter 5 - Packet-Based Estimation and Control}}
 
* Lecture notes: {{ncsbook|fusion|Chapter 5 - Packet-Based Estimation and Control}}
  
====  Additional Information ====
+
== Reading ==
 
 
|}
 
 
 
==== Further Reading ====
 
 
* <p>[http://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=29437 Kalman Filtering with Intermittent Observations], B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, M. Jordan and S. Sastry. ''IEEE T. Automatic Control'', 2004.  This is the paper that covers the main results of this lecture.</p>
 
* <p>[http://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=29437 Kalman Filtering with Intermittent Observations], B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, M. Jordan and S. Sastry. ''IEEE T. Automatic Control'', 2004.  This is the paper that covers the main results of this lecture.</p>
  
 
* <p>[ftp://ftp.cds.caltech.edu/murray/preprints/estm07-automatica_s.pdf Probabilistic Performance of State Estimation Across a Lossy Network].  Michael Epstein, Ling Shi, Abhishek Tiwari and Richard M. Murray.  ''Automatica'',  2008 (to appear).  This article describes how to computer probabilistic guarantees on estimator performance in the presence of packet loss.</p>
 
* <p>[ftp://ftp.cds.caltech.edu/murray/preprints/estm07-automatica_s.pdf Probabilistic Performance of State Estimation Across a Lossy Network].  Michael Epstein, Ling Shi, Abhishek Tiwari and Richard M. Murray.  ''Automatica'',  2008 (to appear).  This article describes how to computer probabilistic guarantees on estimator performance in the presence of packet loss.</p>

Revision as of 01:02, 29 March 2008

Prev: State Estimation Course home Next: Graph Theory

This lecture describes how to extend results in estimation and control to the case where the information between sensing, actuation and computation flows across a network with possible packet loss and time delay. We begin with the estimation problem, summarizing the results on Sinopoli et al on Kalman filtering with intermittent data, which uses average convergence as a stability metric. An alternative formulation is to use almost sure convergence, which gives improved results for lossy networks. Finally, we extend the results on estimation to the control setting, summarizing approaches in the cases where receipt of packets are acknowledge (TCP-like) or not acknowledged (UDP-like).

Lecture Materials

Reading