MARR: A Multi-Agent Reinforcement Resetter for Redirected Walking

Area of Science:

• Virtual Reality (VR) and Human-Computer Interaction
• Reinforcement Learning and Multi-Agent Systems

Background:

• Redirected Walking (RDW) is a technique to navigate virtual environments within limited physical spaces.
• Excessive RDW resets can reduce user immersion and presence.
• Existing RDW optimization methods primarily focus on single-user scenarios.

Purpose of the Study:

• To develop and evaluate a novel reset technique for RDW in multi-user virtual environments.
• To minimize the number of resets while maintaining user immersion.
• To account for both static obstacles and dynamic user movements.

Main Methods:

• Proposed Multi-Agent Reinforcement Resetter (MARR) algorithm.
• Trained MARR using multi-agent reinforcement learning to optimize reset directions.
• Evaluated MARR through simulations and user studies against existing RDW techniques.

Main Results:

• MARR significantly outperformed existing RDW reset techniques in minimizing resets.
• MARR effectively adapted to various environmental contexts, including dynamic multi-user interactions.
• The system learned optimal reset directions tailored to different RDW techniques.

Conclusions:

• MARR is an effective reset technique for RDW in multi-user virtual environments.
• The approach enhances user experience by reducing immersion-breaking resets.
• MARR shows potential for integration with future RDW advancements.