The authors extend previous work that introduced the Golog language. Golog is a language for reasoning about state and change, based on the situation calculus, and implemented as a Prolog interpreter.

The extension enables an agent programmed in the language to reason by using sensor information, as well as by using rules stating how situations will change after actions are performed. This means a closed world assumption is no longer required.

As an example, an agent may reason that a light is on because it was switched on in a past situation, and since then, the agent has not performed any action that would switch it off. In the extension, an agent may reason that a light is switched on because its light sensor tells it this is so. Or, an agent may reason that a light is on because the agent’s light sensor has told it so in the past, and the agent has performed no action since then that would switch the light off.

Golog has the capacity to alternate between online and offline modes during execution. In online execution, commands are executed by real actions and cannot be undone. Offline execution represents planning ahead, so commands can be undone, and non-deterministic alternatives explored. In offline execution, an agent can reason forward from sensor readings it already knows, but obviously will not know the sensor readings of the future.

The formalism in the paper should not be beyond the competence of any computer scientist willing to commit a little time. Familiarity with Prolog will help readers to grasp the parts on implementation.