Human-guided learning of social action selection for robot-assisted therapy
Emmanuel Senft, Paul E. Baxter, Tony Belpaeme · International Conference on Machine Learning · 2015
This paper presents a method for progressively increasing autonomous action selection capabilities in sensitive environments, where random exploration-based learning is not desirable, using guidance provided by a human supervisor. We describe the global framework and a simulation case study based on a scenario in Robot Assisted Therapy for children with Autism Spectrum Disorder. This simulation illustrates the functional features of our proposed approach, and demonstrates how a system following these principles adapts to different interaction contexts while maintaining an appropriate behaviour for the system at all times.