A behavioural paradigm of artificial intelligence and its application to satellite autonomy
Craig A. Lindley · UNSWorks (University of New South Wales, Sydney, Australia) · 2022
This thesis develops a novel approach to autonomous agency that integrates concepts of behavioural artificial intelligence with the ideas of plans-as-communications, represented knowledge as text, and a multi-agent systems view of representation and reasoning. The primary hypothesis of the research is that this integrated approach is a viable basis for high-level satellite autonomy. To investigate this hypothesis, a behavioural control layer is designed that performs the survival function of autonomous power control. A satellite and orbit simulator are developed to test the control model, and simulation runs using different orbit parameters demonstrate controller operation within the simulated system. An adaptive version of the power controller is then developed that automatically generates critical control parameters, and adapts to the natural variation of satellite orbit parameters during the spacecraft lifetime. Test in the simulation environment demonstrate rapid adaptation of the controller to changing orbit characteristics. The control system is further developed to address autonomous planning and scheduling of data operations, based upon the view of plans-as-communication. The satellite simulator is extended to include stochastic command generation. A behavioural control layer is developed that processes using a simple goal arbitration mechanisms. This novel implementation of the concept of plans-as-communications operates within the power constraints generated by the first level power controller. The behavioural planner has the advantage of not depending upon deliberation or world representation, and also provides a prototypical approach to generalised language use by autonomous agents. Within the integrated behavioural approach, logics are regarded as social decision-making conventions, and knowledge representations as socially available texts that can be read by intelligent agents. To begin to explore these ideas, the thesis considers how the behavioural system can participate in joint activities carried out by a group of similar agents. The particular task of a user searching for material through a group of interconnected satellites is considered. A novel protocol is defined for conducting search as a joint activity, and simple modifications to the planning functions of the behavioural satellite are presented that allow the satellite to use the protocol.