Planning with hierarchical task networks in video games
John-Paul Kelly, Adi Botea, Sven Koenig · 2007
Artificial intelligence (AI) technology can have a dramatic impact on the quality of a video game. AI planning methods are useful in a wide range of game components, including modules to control the behaviour of fully autonomous units. However, planning is computationally expensive and the CPU and memory resources available at runtime to a game AI module are scarce. Offline planning can be a good strategy to avoid a runtime performance bottleneck. In this work we apply planning with hierarchical task networks (HTNs) to video games. HTNs can speed up planning dramatically, since search is guided with human-encoded knowledge. We describe an architecture that computes plans offline. This can be seen as a form of generating scripts automatically, replacing the traditional approach of composing them by hand. The results are very encouraging. Scripts are automatically generated at a level of complexity that would require a great human effort to create.