Failure Probability of Verifiable Goal-based Control Programs due to State Estimation Uncertainty
Julia M B Braman, Richard M Murray
Conference on Decision and Control, 2008 (submitted)
Fault tolerance and safety verification of control systems that have state estimation uncertainty are essential for the success of autonomous robotic systems. A software control architecture called Mission Data System, developed at the Jet Propulsion Laboratory, uses goal networks as the control program for autonomous systems. Certain types of goal networks can be converted into linear hybrid systems and verified for safety using existing symbolic model checking software. A process for calculating the probability of failure of some verifiable goal networks due to state estimation uncertainty is presented. Extensions of this procedure to include other types of uncertainties are discussed, and example problems are presented to illustrate these procedures.
- Preprint: http://www.cds.caltech.edu/~murray/preprints/bm08-cdc_s.pdf