Context-sensitive behavior is vital for the success and perhaps even the survival of agents performing complex tasks in the real world. The approach described in this paper is one attempt to give real-world reasoners context-sensitivity. It relies on explicitly representing contexts as contextual schemas, then retrieving and merging them to form a picture of the agent's current context. This ``current c-schema'' is then used to guide virtually all facets of the agent's behavior, assuring behavior that is automatically tailored to the situation. As the situation changes, the current c-schema is updated accordingly, automatically keeping the agent's behavior appropriate for its evolving problem-solving situation.
Though our approach has been and continues to be developed for a particular kind of cognitive architecture (schema-based reasoning), it potentially has a much broader range of applicability. We believe the approach can be used equally well to augment rule-based, blackboard-based, plan-based, and most other kinds of reasoners so that they, too, can be more context-sensitive.
We are currently developing our work in the domain of autonomous underwater vehicle control, both for AUVs operating singly and in groups, as part of the ORCA and MAVIS [\protect\citenameTurner et al., 1991] projects of the UNH Cooperative Distributed Problem Solving research group. Our work will initially be tested via experiments in our simulation testbed [\protect\citenameTurner et al., 1991], but our ultimate target is to evaluate our work aboard AUVs as they undertake useful ocean science missions.