Towards Walkable and Safe Areas: DRL-Based Redirected Walking Leveraging Spatial Walkability Entropy

Area of Science:

• Virtual Reality
• Human-Computer Interaction
• Robotics

Background:

• Redirected walking (RDW) enhances virtual exploration in limited physical spaces.
• Current RDW controllers lack advanced spatial feature extraction for effective obstacle avoidance.

Purpose of the Study:

• To develop a novel spatial walkability-aware redirection controller using deep reinforcement learning (DRL).
• To improve obstacle avoidance capabilities by leveraging comprehensive spatial features and a new walkability metric.

Main Methods:

• Introduced Spatial Walkability Entropy (SWE) to quantify position safety and reachability.
• Designed a joint reward function integrating SWE and virtual-physical alignment.
• Implemented a novel reset strategy to guide users to open areas and minimize re-collisions.

Main Results:

• The proposed DRL controller significantly reduced physical collisions compared to state-of-the-art methods in simulations.
• Live user experiments validated a superior roaming experience and enhanced safety in practical virtual reality settings.

Conclusions:

• The spatial walkability-aware controller offers a significant advancement in RDW for virtual reality.
• This approach enhances user safety and immersion by effectively mitigating physical obstacles.