Goal-oriented hierarchical task networks and its application on interactive narrative planning

Official URL: http://risc01.sabanciuniv.edu/record=b2325823 (Table of contents)

Two of the most commonly used AI architectures in digital games are Behavior Tree (BT) and Goal-Oriented Action Planning (GOAP). The BT architecture is script based, highly controllable but barely expandable. On the other hand the GOAP architecture is planner based, barely controllable but highly expandable. This thesis proposes a hybrid AI architecture called Goal-Oriented Hierarchical Task Network (GHTN); combining planner based approach of GOAP with script based approach of BT. GHTN modifies the Hierarchical Task Network (HTN) architecture by replacing its iterative planner with a goal oriented planner, while maintaining the BT-like scripting capabilities of HTN. GHTN's iterative-planner hybrid architecture is suitable to be used for Interactive Narrative Planning. Using GHTN with a previously crafted domain, it is possible to obtain a non-repetitive and continuous narrative flow which can also be directed by external goals. The user is presented with choices that are intelligently chosen to push the narrative towards the goal; then, depending on the answers new choices are generated. The initial state of the world and the goals are specified by a Scenarist who has the knowledge of the domain. The proposed architecture is tested on Interactive Narrative Planning task with an example domain set in the Lala Land universe, and the architecture is tested with several initial world states and goals.