Research has shown that there is a correlation between learner affect and leaming success. The detrimental effects of unfavorable affect have been observed both in traditional classroom settings as well as computer-enhanced leaming. Consequently, it is of major importance to identify effective Intervention strategies to improve the leaming process. Studies in area of intelligent tutoring Systems have shown that a tutoring System that is sensitive to learner affect may have a positive impact on learning.

The objective of this project is to develop a sensitive tutoring System for children that adapts its pedagogical strategies continuously based on affect-related learner States that are likely to have a negative impact on the learning process.Specifically, we focus on disengagement, which is usually caused by a negative attitude towards the learning object, the learning environment, or the teacher.

The project combines advances in crossmodal machine learning. The tutor's behavior is dynamically adapted to the learner's detected states and their learning progress. For this, causal relationships between the tutor's actions and learner disengagement are investigated to subsequently improve the tutor's actions with reinforcement learning (RL).