Human-robot information sharing with structured language generation from probabilistic beliefs

摘要: This paper presents a framework for information sharing and fusion in cooperative tasks involving humans robots. In this context, all regarding the state of interest is recursively fused maintained by each agent form belief. For robot agent, its belief commonly practically represented as probability density function (pdf), formed traditional sensor estimation algorithms. with non-expert humans, needs to effectively communicate so that gathered can be easily processed interpreted humans. The goal research provide two-way exchange between robots former operating on pdfs, while latter English sentences. achieved considering two goodness measures: semantic correctness preservation. Based measures studied, results show proposed able generate optimal statements describing given pdfs successfully recover initial inputs used them. Additionally, order describe complex Mixture Statements (MoS) model such expression generated through composition more than one statements. With nonparametric Dirichlet Process MoS generation, it found determine correctly number well corresponding reference parameters needed hypotheses underlying