ADAPTING TO LEARNER’S COGNITIVE DIFFERENCES USING REINFORCEMENT LEARNING
Loading...
Date
2023-05-02
Journal Title
Journal ISSN
Volume Title
Publisher
Nazarbayev University School of Engineering and Digital Sciences
Abstract
This report explores the benefits and challenges of
creating adaptive robotic systems by integrating open-source
software and reinforcement learning algorithms. The aim is to
develop robotic systems that adapt to cognitive differences of
the child to increase the engagement time and develop social
skills using robot-assisted therapy. The Furhat and NAO robots
are used as the platforms for receiving inputs and outputs,
while a reinforcement learning agent selects the order of ac-
tivities. The state space includes the pose landmarks from the
Mediapipe and OpenPose libraries to recognize the engagement
level and some static parameters such as age, autism level,
verbality, and the co-diagnosis of Attention Deficit Hyperactivity
Disorder. The actions space contains the act of changing the
current activity. In the current stage, we have eight types of
activities the robot choose to interact with the child. Furhat
robot is programmed to have a full pipeline for a conversational
robot: automatic speech recognition, text-to-speech, machine
translation, and OpenAI language model. The results of this project provide insight into the potential of using open-source software and reinforcement learning algorithms to create more advanced interactive robots and highlight the importance of continued research in the field of Human-Robot Interaction.
Description
Keywords
Type of access: Restricted, reinforcement learning, child-robot interaction, robot-assisted autism therapy
Citation
Issa, I., Kassymbekov, S., Kuangaliyev, Z., & Nurgazy, S. (2023). Adapting to Learner’s Cognitive Differences Using Reinforcement Learning. Nazarbayev University School of Engineering and Digital Sciences