Efficient intent-based narrative generation using multiple planning agents

In Interactive Storytelling (IS) the prevailing approach for the automatic generation of plausible narratives that meet global author goals is intentional planning. However, existing approaches suffer from limited expressiveness and poor scalability. We address this by replacing single intentional planners with multiple agents representing the characters of a narrative, which can reason about the relevance of narrative actions given their individual intents. These are then combined using a state-based forward search procedure that results in a significantly smaller search space.

Unlike other multiagent approaches, these agents calculate all reasonable plans in a state. This allows a search of a wide range of narrative possibilities prior to execution as in planner-based approaches, rather than agents making early plan commitments in a simulation.

We demonstrate that this not only produces the same forms of narrative as single intentional planners but can be extended to generate narratives that are beyond their scope. We also present a search heuristic that exploits the agents' relevant actions to further reduce the size of the explored search space. Experimental results demonstrate system performance that makes it suitable for use in real-time applications such as IS.