A Quantised State Systems Approach Towards Declarative Autonomous Control Ph.D. Thesis

This thesis concerns methods for developing control software for autonomous systems with a high level of modularity, striving towards a declarative control paradigm, where the control system is able to solve control and estimations tasks based on system and objective descriptions alone. This is pursued on the basis of Quantised State Systems (QSS) which is a recent formalism for working with systems of ordinary differential equations in an computing environment based on discrete interactions between model entities. At first; Quantised State Systems are introduced together with the Discrete EVent Systems (DEVS) formalism used to implement them. A comparative study on simulation performance compared to traditional time-discrete methods is given for an autonomous underwater vehicle. Next, a novel Extended Kalman Filter (EKF) variation is developed which utilises the QSS approach to allow Jacobian free estimation with discrete event inputs. This approach is compared to a traditional EKF implementation on an example concerning attitude determination for a deep-space-probe. Two different control strategies are thereafter developed based on the QSS approach; an optimising general controller that uses local information to provide an input signal that minimises a user supplied objective function of the state, and a controller based on sliding mode control which is highly modular and also allows configuration by supplying an objective function. The aforementioned QSS based estimator and control algorithms are evaluated in a closed-loop control setting with a high-fidelity simulation model simulating a Deep Space Probe conducting a Jovian fly-by. The results favours the sliding mode controller in terms of both performance and robustness. A simulation architecture for Hybrid System models are developed, allowing translation from XML specifications of hybrid models into run-time representations. It is demonstrated how the tools developed for hybrid model simulation can be combined with the aforementioned algorithms for continuous control to implement hybrid supervisory control systems. Finally, it is proposed how the different algorithms and tools can be combined into a declarative control system. Synopsis Danish Abstract Nærliggende afhandling omhandler metoder til udvikling af software til kontrol at autonome systemer med en høj grad af modularitet. Udviklingen sigter imod et deklarativt kontrol paradigme, hvor kontrol systemet bliver i stand til løse estimationsog kontrolopgaver baseret på modelbeskrivelser og målbeskrivelser. Denne målsætning forfølges på basis af Kvantiserede State Systemer (KSS), som er en ny formalisme for behandling af ordinære differentialligninger i et computersystem baseret på diskrete interaktioner imellem komponenter. Indledningsvis beskrives Kvantiserede State Systemer sammen med en specifikation af Discrete EVent Systems (DEVS), som er en formalisme der implementerer KSS systemer. Et simuleringstudie sammenligner KSS simulering med traditionelle tidsdiskrete metoder for et eksempel omhandlende en autonom undervandsbåd. Derefter udvikles en ny variation af den kendte Extended Kalman Filter (EKF) algoritme baseret på KSS systemer. Denne tilgangsvinkel tillader estimering uden kendskab til Jacobian matricen for systemet og understøtter event baserede målinger. Den nye algoritme sammenlignes med den traditionelle EKF algoritme på et eksempel omhandlende en interplanetarisk probe. To forskellige reguleringsalgoritmer er herefter udviklet med base i KSS tilgangen; en optimerende regulator, som benytter lokal information til at udlede et styresignal der minimerer en brugerdefineret målfunktion, og en modulær opbygget controller baseret på teorien om glidende manifolder, som også styres af en brugerdefineret målfunktion. De førnævnte KSS baserede værktøjer til estimering og kontrol er evalueret på et reguleringsproblem omhandlende en interplaneterisk probe der flyver forbi Jupiter. En simuleringsarkitektur for modeller af hybride systemer er udviklet, som tillader oversættelse fra XML specifikationer til software-objekter. Det er demonstreret hvordan disse værktøjer for hybrid simulering kan kombineres med førnævnte algoritmer og implementere hybrid control systemer. Endeligt, er det forelsået hvordan resultaterne kan videreudvikles henimod et deklarativt kontrol system. Acknowledgements I would like to thank my two supervisors associate professor Jan Dimon Bendtsen and professor Jakob Stoustrup for continuous support and input during the past three-and-a-half years. I have been privileged to pursue my own ideas to a great extend during this work and appreciate the patience and trust granted me in so doing. I hope we will continue to work as partners on joint projects in the future. A special thanks to Jan Dimon Bendtsen and Karl Kaas Laursen for the fruitful discussions of our ad-hoc discussion group about hybrid systems and Java technology during 2005. I would also like to thank professor Kristin Ytterstad Petersen for hosting me at the Norwegian University of Science and Technology during the fall of 2006. It was a pleasant period and opened my eyes to the fascinating world of underwater robotics. Thanks are also due to all my colleagues at the section of Automation and Control at Aalborg university with whom I have shared many exciting experiences over the years satellite launches not least. A special thanks to Karen Drescher for help tackling the bureaucracy seemingly inherent to life at the university. A very special thanks is also due to Morten Bisgaard with whom I have had an excellent cooperation over the last past 8 years. In the same paragraph I wish to thank Tor Viscor for friendship over the years. Finally, the greatest thanks goes to Birgit Krogh for love and support, and for clearing the clouds when it all looked almost impossible. A special thanks for the patience you showed during the last months.