Enhancing Free Walking in Virtual Environments with Warning Walls: A Pilot Study on Redirected Walking Using Machine Learning Agents

Abstract

Virtual Reality (VR) technology allows users to explore unlimited virtual environment spaces. Users can explore virtual spaces using a Head Mounted Display (HMD) device. Various techniques for exploring VR spaces, such as teleportation, flying, walking-in-place, and devices such as omnidirectional treadmills, are often used. Previous literature states that real walking is the most natural method to explore virtual spaces. However, the constraint to the natural walking method is the limited physical space, so the user is at risk of encountering the boundary wall of physical space. Redirected Walking (RDW) technique addressed solving the limitations of tracking space so that users can explore unlimited virtual spaces without encountering the boundaries of physical space. The RDW technique has experienced rapid development since it was first introduced more than two decades ago; until now, its development has utilized artificial intelligence technology. This pilot study aims to explore the use of AI in the Deep Reinforcement Learning framework in designing virtual environment designs through a simulation system. This study uses machine learning agents trained to explore a non-predefined pathway virtual space larger than the tracking space by applying the essential principles of redirected walking techniques such as rotation, translation, and curvature gain. The study results show that machine learning agents can learn well and explore virtual spaces larger than the size of tracking spaces, both using warning walls and without warning walls.

Keywords: Deep reinforcement learning, machine learning agents, redirected walking, virtual environment.

DOI: 10.54941/ahfe1006342