Controlling Semi-automatic Systems with FLUX

The programming of agents that are capable of reasoning about their actions is a major application of logic programming in Artificial Intelligence. FLUX is a recent, constraint logic programming-based method for designing reasoning agents that can sense, act, and plan under incomplete information [3]. An agent program in this language consists of a background theory, which endows the agent with the necessary knowledge of its actions, along with a high-level acting and planning strategy. The reasoning facilities are provided by a constraint solver, which is formally based on the action theory of the fluent calculus and its solution to the frame problem under incomplete states. We have extended FLUX by a method that allows agents to reason, plan and act in semi-automatic environments, in which actions can initiate whole chains of indirect or delayed effects. Our approach addresses complex issues such as simultaneous, additive changes of state variables under incomplete information. As a case study, we have developed a control program for a complex dynamic environment of a steam boiler. A model of a real system, the domain involves uncertainty in form of varying output of the water pumps, which requires the agent to reason and plan under incomplete information. Moreover, simple actions in this system, such as turning off a pump, may trigger a whole chain of indirect effects: Another pump may automatically change to high capacity, which in turn may cause a valve to open, etc. Furthermore, every one of these changes has a specific effect on the water level in the boiler, which in turn affects the quantity of steam that is produced. As an innovative and versatile way of modelling additive fluents and delayed effects, we introduce the notion of a momentum fluent. Such a fluent describes a property of the momentum in a physical system rather than a static property. An action may cause several momentum fluents to become true, each of which represents a specific contribution to the same fluent (such as the water level in the steam boiler). Additive fluents can thus be modelled. In a similar fashion, an action which has a delayed effect can be specified as bringing about a momentum fluent that eventually causes the actual delayed effect. Having triggered this effect, the momentum fluent itself may either automatically terminate, or continue to hold in case of recurring effects. On the other hand, the agent may have at its disposal …