Abstract
Goals and policies are presented as special kinds of rules for managing systems in a flexible way without requiring specialised technical knowledge. However, it is not always possible to rely on exact information for such an approach. Policies should therefore not have to be formulated in terms of precise inputs and outputs. Instead, it is desirable to allow loose goals and policies that accommodate probabilistic system inputs/outputs and fuzziness in rules. This is a general solution that is relevant to many different kinds of applications. The paper uses automated home care management as a concrete illustration of how the approach works. The overall system architecture is presented, along with an overview of the language for expressing goals and policies. The extensions made to allow looser formulations are described. An extended worked example explains various aspects of the approach. The paper concludes with a user evaluation and a discussion of the work.