Cyril Brom
brom [--at--] ksvi. -tobedeleted- mff.cuni. cz

Action Selection for Virtual Humans in Large Environments

Abstract    Downloads

This website is devoted to our approach to action selection of virtual humans acting in large environments. Specifically, our method features level-of-detail for artificial intelligence and virtual space. The approach is descibed in the doctoral thesis "Cyril Brom: Action Selection for Virtual Humans in Large Environments" (in Czech).

On this website, you can obtain the current versions of prototypes implementations, namely Project IVE and Project ENTs, and general information.

Key words: virtual human, large virtual environment, action selection, reactive planning, intention, virtual affordance, level of detail AI, role-passing

Abstract of the thesis

This thesis presents ISMA a representation of procedural knowledge of virtual humans and S-GHRP, an algorithm for controlling virtual humans exploiting this representation. ISMA and S-GHRP departs from similar approaches in that they are aimed at virtual humans with complex behaviour acting in large environments. Specifically, ISMA and S-GHRP cope with (1) supplying the simulation in run-time with new components (i.e., objects, actions, places), but without necessity of using any machine learning mechanism at the side of virtual humans, (2) simplifying the simulation at the places out of centre of attention using a smooth level-of detail technique for behaviour and space, (3) supporting transition behaviours.

ISMA and S-GHRP are based on the perceptual theory of affordances of J. J. Gibson, and practical reasoning framework of M. E. Bratman. Actually, ISMA and S-GHRP present their implementable refinement. Besides presenting the conceptual description of ISMA and S-GHRP, two particular prototype implementations are described.

The extended abstract is also available (in English).


The thesis can be downloaded here... (in Czech).

The principles detailed in the thesis are demonstrated in projects Project IVE and Project ENTs, which can be downloaded from their web sites. The related project Pogamut using some of this principles as well, is also available.

Besides the projects, a video demonstrating capabilities of a virtual gardener driven by our reactive planning method can be also downloaded (70 MB).

Last updated: 4.1.2007.