A Model Driven Approach for Eased Knowledge Storage and Retrieval in Interactive HRI Systems

Abstract

Efficient storage and querying of long-term human-robot interaction data requires application developers to have an in-depth understanding of the involved domains. It is an error prone task that can immensely impact the interaction experience of humans with robots and artificial agents. In the development cycle of HRI applications, queries towards storage solutions are often created once, copied into according components, and are rarely revisited. Beyond possible syntactical errors (especially impacting query design time), any change in the underlying storage solution structure results in semantic errors at run time which are not easy to spot in existing applications. To address this issue, we present a model-driven software development approach to create a long-term storage system to be used in highly interactive HRI scenarios. We created multiple domain specific languages that allow us to model the domain and seamlessly embed its concepts into a query language. Along with corresponding model-to-model and model-to-text transformations we generate a fully integrated workbench facilitating data storage and retrieval. It supports developers in the query design process and allows in-tool query execution without the need to have prior in-depth knowledge of the domain. We evaluated our work in an extensive user study and can show that the generated tool yields multiple advantages compared to the usual query design approach.