Reasoning with Unreliable Observations
in the Features and Fluent Approach

Silvia Coradeschi

Dept. of Computer and Information Science, Linköping University, Sweden


In this work we consider a way to deal with the problem of unreliable observations when an agent is reasoning about dynamical systems as they are formalized and systematically studied in Sandewall's approach to reasoning about action and change. The presence of incorrect observations can be detected in case it generates a contradiction. In this case a revision function resolves the inconsistency by constructing alternative consistent descriptions of the system. However, revision is in general expensive and therefore we define a delayed revision that postpones doing revision for some steps and we prove that it gives results similar to immediate revision. We also examine some forms of preferential revision that reduce the number of alternative descriptions of the system. We finally consider the relations between our work and Gardenfors' approach to belief revision.

Download the paper (126691 bytes).

Silvia Coradeschi <>