Abstrakt
Episodic memory modeling enjoys increasing interest in virtual agents, autonomous companions or computer games research communities. To evaluate memory models, it is often necessary to \fill them with appropriate data - streams of activities corresponding to what would happen to a real human during a particular time period.
However, such activity corpora, freely available, are, to our best knowledge, lacking. This paper has two goals.
First, it shows two already implemented complementary approaches to generating activity corpora. While the first one uses HTN planning to create a corpus with relatively abstract activities, the second one utilizes a 3D simulation to generate a stream of more fine-grained actions.
The key question is evaluation of the resulting corpora in terms of their resemblance to streams of activities of real humans. Thus, the second goal of this paper is to compare the generated corpora to several known datasets of human activity, based on their entropy, compressibility and statistics of transitions between actions.
This can be conceived as a first step towards creating general complexity measures of streams of human actions and we see it as the main contribution of this paper.