The importance does not lie in the existence of a single plan, but rather in a system's ability to predictively manage plan execution in light of continuous feedback from an environment and to replan when failures occur.
Entropy Reduction Engin is an architecture for the intergration of planning, scheduling (sequencing a given set of actions in terms of metric time and resource constraint), and control. The architecture is motivated, presented, and analyzed in terms of its different components: namely, problem reduction, temporal projection, and situated control rule execution.
A user must provide a causal theory for the domain of application, or no behaviors can be produced by the system. This theory consists of a description of the control actions that can be taken by the system, their predictions, and probability distributions of these actions' possible effects. In a simiar fashion, a user can also specify exogenous events that are outside of the system's control. This information is used by the Projector to reason about possible system behaviors. To complete the causal theory, the user must provide domain constraints that specify those facts which can never co-occur.
A user can (but need not) provide a statement of the goal that is to be satisfied by the system which are called brhavioral constraints. A behavioral constraint is used by the Projector to evaluate possible system behaviors. The projector builds a graph of various possible future behaviors and incrementally evaluates the given behavioral constraint with respect to this graph. Once a satisfactory graph is found, it is compiled into situated control rules (SCRs) that are used by the Reactor to determine which action to take in a given situation. In addition the Reactor uses the behavioral constraint to determine whether it has satisfied the user's overall goal by incrementally comparing current sensor values against the values given in the constraints.
A user can provide (but need not) provide a set of production rules. These rules specify how to reduce (or rewrite) a problem into a conjunction of subproblems.
The operator representations allows for actions and events to have various outcomes associated with probabilities of occurence.
The Projector and Reactor have been designed to have anytime characteristics: the amount of knowledge on which the synthesized advice is based increases with the amount of available computation time.
Back to the Title Page
Architecture
The ERE architecture consists of three major components: the Reactor, the Projector, and the Reactor. The Reductor synthesizes appropriate problem solving strategies for a given problem; the Projector uses these strategies as search control to plan and schedule appropriate actions; and the Reactor executes control rules derived from the Projector's plans.
Agent Properties
The Entropy Reduction Engin (ERE) project is a focus for research on planning and scheduling in the contexts of closed-loop plan execution where a plan describes a desired behavior, and feedback from the environment is used to measure deviations.
Capabilities
Environment and Agent Body
Issues
Other Architectures