Extension of novel model for robots to learn from communication: English verbalizations of its logical forms
Open Access
Author:
Liu, Aishan
Area of Honors:
Computer Science
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Rebecca Jane Passonneau, Thesis Supervisor Ting He, Thesis Honors Advisor
Keywords:
Natural Language Processing Learning from Demonstration meaning representation language data augmentation dialogue manager Quarto Krippendorff's alpha
Abstract:
Robots that can actively engage in board games and learn game strategy could contribute to the field of socially assistive robotics, by playing a role in education and elderly person caring. To achieve such benefits, we present a robot that engages in short questioning dialogues to learn board games like Connect Four, Gobblet, Quarto. Learning the game means learning to add paths to a game tree representation, that can then be used as input to an algorithm that searches the game tree during play with a human.
Robot Learning from Demonstration (LfD) is a paradigm for enabling robots to learn by communicating with humans. The novel method applied to our robot learning model is combining a dialogue manager with the game-theoretic learning from demonstration(LfD) approach to ask verbal or visual questions about win conditions of the games, which are then added to its knowledge as paths in its game-tree representation. The communicative actions in the dialogue are represented in a formal meaning representation language. This is the language the agent uses in interactions with a simulated dialogue partner during off-line reinforcement learning of dialogue policies. As a contribution to our robot project, this thesis focuses on extending on our novel robot learning model with English verbalizations of its logical forms.
