This is the implementation of an accepted short paper for the Conference on Spatial Information Theory (COSIT'22).
People can be localized at a particular location in an indoor environment using verbal descriptions referring to distinct visible objects (e.g., landmarks). When a user provides an incomplete initial location description their location may remain ambiguous. Here, we consider a dialogue initiated to update the initial description, which continues until the updated description can be related to a location in the environment. In each interaction, the wayfinder is incrementally asked about the visibility of a particular object to update the initial description. This paper presents an entropy-based model to minimize the number of interactions. We show how this entropy-based model leads to a significant reduction of interactions (i.e., reduction of conversation length, measured by the number of additional referents) compared to baseline models. Moreover, the effect of the initial description, i.e., the first set of visible objects with different combinations, is investigated.
python conversation.pypython result_analysis.py
python AVG_Comparison.py