ENI: Quantifying Environment Compatibility for Natural Walking in Virtual Reality

Niall L. Williams, Aniket Bera, Dinesh Manocha

Introduction

Natural walking—physically moving one's legs to traverse a space—is considered the "gold standard" for immersion in Virtual Reality (VR). However, a significant gap exists between the expansive virtual environments (VE) designers create and the small, obstacle-filled physical environments (PE) where users actually move. While techniques like Redirected Walking (RDW) attempt to bridge this gap by subtly manipulating a user's path, there has historically been no formal way to quantify if a specific VE is even navigable within a given PE.

The Environment Navigation Incompatibility (ENI) metric addresses this by providing a general framework to quantify how well a physical room "fits" a virtual world. By comparing local geometric visibility, ENI predicts navigation difficulty and collision risks without requiring prior user movement data.

Methods: A Geometric Correspondence Problem

The authors treat navigability as a shape correspondence problem, calculating compatibility through a multi-step geometric pipeline:

  1. Uniform Sampling: The free space of both the PE and VE is sampled using a Conforming Constrained Delaunay Triangulation (CDT) to identify representative points.
  2. Visibility Polygons (Isovists): At each point, the algorithm computes a visibility polygon—the 2D area a user can theoretically walk toward without hitting an obstacle. This captures the "local neighborhood" of a location.
  3. The ENI Metric: For every point in the virtual world, the algorithm searches the physical world for the most compatible location. Compatibility is defined by the Boolean difference between the two visibility polygons. If a large portion of the virtual walkable area is obstructed in the physical counterpart, the incompatibility score increases.
  4. Optimization: To account for user orientation, the metric rotates the polygons to find the alignment that minimizes the mismatch.
Comparison of physical and virtual free spaces to determine local compatibility.

The output is an $n$-dimensional vector (one score per virtual point), typically summarized by its mean ($\mu$) and standard deviation ($\sigma$).

Key Insights: The Clutter Paradox

A counter-intuitive finding of this research is the "Clutter Paradox": adding obstacles to a virtual environment can actually improve its compatibility with a small physical room. In a wide-open VE, users tend to walk in long, straight lines, quickly reaching physical boundaries. Conversely, a cluttered VE forces frequent turns, which naturally keeps the user within the bounds of a restricted physical space.

Furthermore, ENI allows designers to visualize incompatibility by heat-mapping scores onto a VE map. This enables the identification of "danger zones"—specific corners or hallways mathematically prone to causing physical collisions.

Results and Validation

The metric was validated using synthetic environments, RRT* path simulations, and user studies involving 30 participants.

Heat-mapped ENI scores for three virtual environments (VE) relative to a single physical environment (PE).

Key findings include:

  • Strong Correlation: Simulations showed that as ENI scores increased, the average distance users could walk before a "reset" (hitting a physical wall) significantly decreased.
  • Predictive Power: In user studies, the metric successfully predicted that larger VEs would lead to more resets. For instance, moving from a 400m² VE to a 1600m² VE caused the mean ENI to jump from ~80 to ~760, while the distance between resets dropped accordingly.
  • RDW Evaluation: ENI proved effective at evaluating RDW controllers. An Alignment-based controller (ARC) produced paths with significantly lower ENI scores ($\mu=11.4$) compared to standard Steer-to-Center algorithms ($\mu=34.1$), explaining why ARC provides a more seamless experience.

Limitations and Future Work

Currently, ENI is limited to static environments and does not account for dynamic obstacles like moving people or pets. The exhaustive search for "best-fit" points is also computationally expensive in its current Python implementation. Future work aims to extend the metric to 3D spaces (accounting for slopes and ceilings) and incorporate temporal components to handle dynamic environments. Despite these limitations, ENI represents a significant step toward treating the physical and virtual worlds as a single, coupled system for immersive navigation.

Summary updated

BibTeX
@inproceedings{williams2022eni,
  author = {Niall L. Williams and Aniket Bera and Dinesh Manocha},
  title = {{ENI: Quantifying Environment Compatibility for Natural Walking in Virtual Reality}},
  booktitle = {VR},
  year = {2022},
  url = {https://openreview.net/forum?id=VEY2c-HaTlm},
}
Figure