Asynchronous Knowledge with Hidden Actions in the Situation Calculus

We present a powerful new account of multi-agent knowledge in the situation calculus and an automated reasoning procedure for knowledge queries. Ex-isting accounts of epistemic reasoning in the situation calculus require that whenever an action occurs, all agents know that an action has occurred. This demands a level of synchronicity that is unreasonable in many multi-agent domains. In asynchronous domains, each agent’s knowledge must instead account for arbitrarily-long sequences of hidden actions. By using a persis-tence condition meta-operator to augment traditional regression techniques, we show how agents can reason about their own knowledge using only their internal history of observations, rather than requiring a full history of the world. The result is a more robust and flexible account of knowledge in the situation calculus suitable for asynchronous, partially-observable multi-agen