Adaptive Virtual Reality Learning Environment with a Reinforcement Learning-Driven Pedagogical Agent

With the fast evolving advances in technology to create immersive learner experiences in Virtual Reality (VR), there are significant opportunities for developing immersive and engaging VR Learning Environments (VRLEs) to train neurodiverse individuals. Integrating Pedagogical Agents (PAs) in these VRLEs has the potential for supporting neurodiverse learners by providing personalized, adaptive guidance tailored to their unique needs. However, a key challenge lies in ensuring the PAs can adapt effectively to the diverse circumstances of the learners. In this paper, we present a novel VRLE viz., uSucceed that leverages a Reinforcement Learning (RL)-driven PA to provide adaptive, personalized support to enhance the training experiences by tailoring the VRLE to the needs of neurodiverse learners. By designing the RL-driven PA algorithm to use Deep Q-Network, we study its capability in guiding the actions of the student in an ongoing learning basis. Simulation experiment results of the speed, accuracy and efficiency of the RL-driven PA in a VRLE demonstrate the benefits of our approach. These results build a strong base for future dynamic VRLE research that can be programmed to adapt based on the learners' experience.