Published on November 26, 2018 | Updated on December 9, 2019


Development of an App for a Lego social robot decoding social intentions

For successful child-social robot interactions, it is crucial that the social robot algorithm allows to infer the child’s intentions and responds accordingly. The aim is to develop a new App for iPhone, iPad and Android based on Partially Observable Markov Decision Processes (POMDP) that help Lego social robots understand the cognitive states of children. Current state-of-the art approaches use inference models as passive processes of observing the behavior of people. Yet, these approaches are missing how intentions of the partner are embedded in a greater interaction context. POMDP is a framework based on probabilistic reasoning about the hidden state of the environment. Based on its observations from the environment, POMDP develops a belief (posterior probability distribution) over the current state and capture the intentions of a human partner based on the context (choice of the human partner in a card game). This allowed us to accurately estimate intentions in a peer-to-computer interaction. We will apply such computer algorithm using POMDP to develop a new App for Lego social robots to infer intentions of children when playing simple games.