The Sabre Narrative Planner
Sabre is a fast multi-agent forward-chaining state-space heuristic search planner used to generate stories. It is a single "puppetmaster" agent that leverages it omniscience to coordinate many characters to ensure they behave realistically according to their (possibly wrong) beliefs and intentions. Sabre ensures that some system-level author goal is achieved, but it only uses actions that every character believes can lead to achieving their individual goals.
Sabre is the successor to Glaive. It supports the same features, plus more, and is significantly more efficient in both memory and time.
The Sabre planner:
- Uses a fluent/value state representations similar to Helmert's Fast Downward planner.
- Uses Riedl and Young's model of intentionality to reason about individual agent goals and cooperation.
- Uses Shirvani, Ware, and Farrell's model of beliefs and theory of mind to reason about what each agent knows and how those beliefs change as a result of actions and observations.
- Uses Ware and Young's model of conflict to reason about how agents plans can interfere with one another and fail. By treating the problem's search space as a set of possible worlds, Sabre can find an explanation for the actions of a failed plan by reasoning about other ways the story might have played out.
- Pure Java 14 (no additional libraries required)
- Rich action syntax, including full ADL support (negative preconditions, typed objects, equality, conditional effects, first-order quantifiers)
- Numeric fluents and arithmetic
- Trigger events, which must occur when they can, and which are often used for belief updates based on observations
- Sabre: A Narrative Planner Supporting Intention and Deep Theory of Mind. In Proceedings of the 17th AAAI International Conference on Artificial Intelligence and Interactive Digital Entertainment, pp. 99-106, 2021. (nominated for Best Paper).
- The Sabre narrative planner: multi-agent coordination with intentions and beliefs. In Proceedings of the 20th International Conference on Autonomous Agents and Multiagent Systems, pp. 1698-1700, 2021..