1. Bibliographic Information

1.1. Title

15 Years of Research on Redirected Walking in Immersive Virtual Environments

1.2. Authors

• Niels Christian Nilsson (Aalborg University)

• Tabitha Peck (Davidson College)

• Gerd Bruder (University of Central Florida)

• Eric Hodgson (Miami University, Ohio)

• Stefania Serafin (Aalborg University)

• Mary Whitton (University of North Carolina at Chapel Hill)

• Evan Suma Rosenberg (University of Southern California)

• Frank Steinicke (University of Hamburg)

  Note: The authors are a consortium of leading experts in Virtual Reality (VR) locomotion and perception from the US and Europe.

1.3. Journal/Conference

IEEE Computer Graphics and Applications

• Status: This is a highly reputable peer-reviewed journal published by the IEEE Computer Society, focusing on the theory and practice of computer graphics and user interfaces.
• Publication Date: January/February 2018 (Online date: 2018-01-12).

1.4. Abstract

This paper serves as a comprehensive survey and review of "Redirected Walking" (RDW), a technique introduced 15 years prior (around 2001) to solve the problem of limited physical space in Virtual Reality. The abstract highlights that while locomotion (moving from place to place) is fundamental to VR, physical boundaries (walls) restrict users. RDW solves this by imperceptibly manipulating the mapping between physical and virtual movements, allowing users to feel like they are walking infinite distances while actually remaining within a confined physical tracking space. The paper reviews techniques, algorithms (controllers), side effects (like simulator sickness), and future challenges.

1.5. Original Source Link

Paper PDF

2. Executive Summary

2.1. Background & Motivation

• The Problem: In Virtual Reality (VR), "Real Walking" (physically walking to move in the virtual world) is considered the gold standard for locomotion because it provides accurate vestibular (balance) and proprioceptive (body position) cues, leading to higher presence. However, it is severely limited by the size of the real-world room (the tracking space). If the virtual world is larger than the real room, the user will eventually hit a physical wall.
• Existing Solutions & Gaps: Alternative methods like "Walking-in-Place" or using joysticks/teleportation solve the space issue but sacrifice realism and can cause motion sickness due to the sensory mismatch (eyes see motion, body feels none). Omnidirectional treadmills are expensive and mechanically complex.
• The Need: There is a need for a software-based solution that allows unconstrained physical walking in large virtual environments (VEs) within small physical spaces without the user noticing the manipulation.

2.2. Main Contributions & Findings

• Comprehensive Taxonomy: The paper categorizes 15 years of RDW research into two main approaches:
  1. Manipulating Gains: Altering the ratio between physical and virtual motion (e.g., rotating the virtual world slightly faster than the user turns their head).
  2. Manipulating Architecture: Changing the virtual world's layout dynamically (e.g., overlapping rooms) to compress space.
• Controller Classification: It identifies three types of algorithms that decide how to redirect users: Scripted (pre-defined), Generalized (reactive, e.g., steer-to-center), and Predictive (guessing user intent).
• Perceptual Thresholds: It compiles key empirical data on "how much" we can trick the human brain before it notices. For example, users can be made to turn physically about 49% more than they think they are turning virtually without realizing it.
• Safety & Side Effects: The review concludes that while RDW is effective, it requires safety mechanisms (overt interventions) when users get too close to walls, and it can increase cognitive load, though it generally preserves spatial memory better than non-walking techniques.

3. Prerequisite Knowledge & Related Work

3.1. Foundational Concepts

• Virtual Reality (VR) Locomotion: The method by which a user moves through a virtual environment.
• Head-Mounted Display (HMD): A device worn on the head (like the Oculus Rift or HTC Vive) that displays the virtual world.
• Tracking Space: The bounded physical area (e.g., a $4m \times 4m$ lab space) where the VR system can track the user's position.
• Vestibular System: The sensory system in the inner ear that contributes to the sense of balance and spatial orientation. It detects rotation and acceleration.
• Proprioception: The body's ability to sense its own position and movement (e.g., feeling your legs move).
• Sensory Conflict Theory: The theory that motion sickness (simulator sickness) arises when there is a mismatch between visual inputs and vestibular/proprioceptive inputs (e.g., seeing yourself move forward but feeling stationary).
• Change Blindness: A perceptual phenomenon where a change in a visual stimulus is introduced and the observer does not notice it.

3.2. Previous Works

• Razzaque et al. (2001) - "Redirected Walking": This is the seminal work cited throughout the paper. Razzaque was the first to practically demonstrate that you can rotate the virtual scene slightly differently than the user's head rotation to steer them away from physical walls.
  • Core Concept: If a user turns their head $90^\circ$ but the scene rotates $80^\circ$, the user will unconsciously turn their body an extra $10^\circ$ to compensate, thus redirecting their physical path.
• Steinicke et al. (2010): A critical study referenced for establishing Detection Thresholds. They used psychophysical methods to quantify exactly how much gain can be applied before a human notices.
• Suma et al. (2011, 2012): Introduced Architectural Manipulation, specifically "Impossible Spaces," showing that users don't notice if virtual rooms overlap physically if the layout changes behind their back or through corridors.

3.3. Technological Evolution

1. Early Days (2001-2007): Focus on Rotation Gains. Simple steering to keep users away from walls.
2. Middle Era (2008-2012): Introduction of Translation and Curvature Gains. Research into Overt Interventions (what to do when redirection fails) like "Freeze-Turn".
3. Modern Era (2012-2017): Emergence of Architectural Manipulation (changing the world itself, not just the motion). Development of advanced Controllers (Predictive algorithms) and investigation into Multi-user redirection.

3.4. Differentiation Analysis

This paper is not a single experimental study but a survey. Its innovation lies in synthesizing disjointed research into a coherent framework. It differentiates techniques into Subtle (imperceptible, ideal) vs. Overt (noticeable, used for safety), and Gain-based (motion mapping) vs. Architecture-based (environment mapping).

4. Methodology

Since this is a survey paper, the "Methodology" section here analyzes the Redirected Walking (RDW) Techniques described by the authors. These are the methods used by researchers in the field to achieve the illusion of infinite walking.

4.1. Principles of Redirection

The core principle of RDW is Perceptual Imperceptibility. The human brain relies on visual, vestibular, and proprioceptive cues to determine motion. However, these senses are imprecise.

• Visual Dominance: Vision often overrides other senses.
• Sensory Thresholds: If the discrepancy between the visual stimulus (what you see) and the vestibular stimulus (what you feel) is small enough (below a detection threshold), the brain ignores the conflict and accepts the visual reality. RDW exploits this "gap" to manipulate the user's physical trajectory.

4.2. Approach 1: Manipulation of Gains (Motion Mapping)

This approach modifies the input-output ratio of user movement. If the user moves by $x$ in the real world, they move by f(x) in the virtual world.

4.2.1. Rotation Gains

Used when the user rotates their head (yaw). The system injects extra rotation into the virtual camera.

• Concept: If a user rotates their head physically by $\alpha_{real}$, the virtual camera rotates by $\alpha_{virtual}$.
• Gain Definition: The rotation gain $g_{rot}$ can be conceptually defined as: $g_{rot} = \frac{\alpha_{virtual}}{\alpha_{real}}$
• Mechanism:
  • If $g_{rot} > 1$ (e.g., 1.2): The virtual world turns faster than the head. To see a target at $90^\circ$ virtual, the user only turns their head $\frac{90}{1.2} = 75^\circ$ physical.
  • If $g_{rot} < 1$ (e.g., 0.8): The virtual world turns slower. To see a target at $90^\circ$ virtual, the user must turn $\frac{90}{0.8} = 112.5^\circ$ physical.
• Goal: This forces the user to reorient themselves physically without noticing, effectively rotating their physical facing direction away from a wall.

4.2.2. Translation Gains

Used when the user walks in a straight line.

• Concept: Scales the distance covered.
• Gain Definition: $g_{trans} = \frac{Distance_{virtual}}{Distance_{real}}$
• Mechanism:
  • If $g_{trans} > 1$: One physical step covers more virtual ground. This allows a large virtual room to fit in a smaller physical dimension.
  • If $g_{trans} < 1$: The user must walk further physically to cover a short virtual distance.

4.2.3. Curvature Gains

Used when the user is walking forward in the virtual world.

• Mechanism: The system slowly rotates the virtual environment (camera) continuously as the user walks forward.
• Effect: To stay on a straight virtual path, the user unconsciously steers their body in the opposite direction of the rotation to compensate.
• Result: The user feels they are walking a straight line in VR, but they are actually walking a circle in the real world. This is the primary method for "infinite" walking (walking in circles within the lab).

4.2.4. Bending Gains

This is a variation of curvature gains applied when the user is already walking a curve in VR.

• Mechanism: It increases or decreases the radius of the curve the user is walking. For example, a slight curve in VR can be mapped to a tighter circle in the real world.

  The following figure (Figure 1 from the original paper) illustrates these four gain types:

  该图像是示意图，展示了四种用于操控用户真实与虚拟运动映射的增益类型：(a) 旋转增益（用户静止）、(b) 平移增益（用户向前移动）、(c) 曲率增益（用户向前移动）和(d) 弯曲增益（用户沿曲线移动）。

• (a) Rotation Gain: User rotates in place; virtual world rotates differently.

• (b) Translation Gain: User walks forward; virtual distance differs.

• (c) Curvature Gain: User walks straight in VR; walks curve in reality.

• (d) Bending Gain: User walks curve in VR; curve radius differs in reality.

4.2.5. Masking Techniques

To hide larger manipulations (gains), researchers use "masking":

• Saccadic Suppression: Injecting redirection during rapid eye movements (saccades) when the brain momentarily "blocks" visual processing.
• Blinks: Shifting the world when the user blinks.
• Visual Distractors / Optic Flow: Using visual noise or peripheral cues to lower sensitivity to the manipulation.

4.3. Approach 2: Overt Redirection (Interventions)

When subtle gains fail (e.g., the user runs at a wall too fast), the system must intervene openly to ensure safety. These are not imperceptible.

1. Freeze-Backup: The VR freezes; the user is told to back up.
2. Freeze-Turn: The VR freezes; the user is told to turn around physically.
3. 2:1 Turn: The user is told to turn around. As they turn $180^\circ$ physically, the system applies a massive rotation gain of 2, so they turn $360^\circ$ virtually.
   • Result: They are physically facing away from the wall, but virtually they are facing their original direction and can continue walking.
4. Distractors: A virtual object (e.g., a butterfly) appears. The user follows it with their head. The system rotates the world while the user is distracted, reorienting them.

4.4. Approach 3: Manipulation of Virtual Architecture

Instead of changing the user's motion, this approach changes the map of the virtual world.

4.4.1. Change Blindness

• Method: The layout of a corridor or the position of a door is changed when it is behind the user's back.
• Example: A user walks down a hall. While they look forward, the door they just came through moves to a different wall. When they return, they walk a different physical path.

4.4.2. Impossible Spaces (Self-Overlapping Architecture)

• Method: Creating virtual rooms that occupy the same physical space but are accessed through different corridors.

• Effect: Figure 2b below shows how two virtual rooms can physically overlap. Because the user cannot see both rooms simultaneously (due to walls), they don't realize they are walking into a space they effectively just occupied.

  The following figure (Figure 2 from the original paper) illustrates these architectural manipulations:

  该图像是示意图，展示了通过虚拟建筑的操控实现的两种重定向形式：(a) 变化失明重定向，用户在空间中行走的过程被调整；(b) 不可能空间，两个房间在物理上重叠，展示了不同的重叠百分比。

• (a) Change Blindness: Notice how the door location moves.

• (b) Impossible Spaces: Two distinct virtual rooms (green and blue) are mapped to overlapping physical space.

4.5. Redirection Controllers

These are the algorithms that decide which gain to apply and when.

1. Scripted Controllers: The path is pre-defined. The developer hard-codes the redirection (e.g., "Apply 1.1 rotation gain at waypoint A"). Good for storytelling, bad for free exploration.
2. Generalized Controllers: Reactive algorithms for free exploration.
   • Steer-to-Center: Always applies rotation/curvature gains to steer the user toward the center of the tracking space.
   • Steer-to-Orbit: Steers the user into a stable circular orbit around the center.
3. Predictive Controllers: Use data (head gaze, environment layout) to predict where the user will go.
   • Strategy: If the system knows the user will walk straight for 10 meters, it can plan a gentle curve. If it predicts a turn, it can prep a rotation gain.
   • FORCE Algorithm: Calculates optimal steering based on a graph of navigable paths.

5. Experimental Setup

This section summarizes the experimental methodologies common in the 15 years of research reviewed by the paper.

5.1. Datasets & Environments

Researchers typically do not use static "datasets" but rather Virtual Environments (VEs) designed for specific tasks:

• Open Fields/Forests: Used for testing infinite walking via curvature gains.
• Corridors/Hallways: Used to test rotation gains at corners or translation gains.
• Virtual Offices: Used to test architectural manipulations (Impossible Spaces).
• Obstacle Courses: To test collision avoidance and steer-to-center algorithms.

5.2. Evaluation Metrics

The paper highlights four key criteria for evaluating RDW techniques:

5.2.1. Detection Thresholds (Imperceptibility)

This measures the point at which a user notices the trick.

• Concept: In a "Two-Alternative Forced Choice" (2AFC) task, a user might turn physically and be asked "Did the virtual movement equal the physical movement?" (Yes/No or Faster/Slower).
• Metric: Point of Subjective Equality (PSE) and Detection Thresholds (typically the 75% detection mark on a psychometric function).
  • Interpretation: If the threshold for rotation gain is 1.1, it means users generally don't notice gains below 1.1.

5.2.2. Simulator Sickness (Side Effects)

• Metric: Simulator Sickness Questionnaire (SSQ).
• Method: Users fill out a survey before and after the experience, rating symptoms like nausea, dizziness, and eyestrain.
• Goal: To ensure RDW does not induce significantly more sickness than standard VR.

5.2.3. Spatial Performance

• Tasks: Pointing to unseen targets, estimating distances, or sketching maps after walking.
• Goal: To verify if RDW distorts the user's mental map of the space. (e.g., "Do I know where the door is after being rotated unconsciously?").

5.2.4. Cognitive Load

• Metric: Dual-task performance.
• Method: Users perform RDW tasks while doing a secondary task (e.g., counting backward by 7s).
• Logic: If RDW requires brain power to process the sensory conflict, performance on the counting task will drop.

5.3. Baselines

RDW is usually compared against:

1. Real Walking (1:1): The gold standard (no manipulation).
2. Walking-in-Place (WIP): Making stepping motions while stationary.
3. Joystick/Controller Locomotion: Standard gaming movement.

6. Results & Analysis

6.1. Core Results: Detection Thresholds

The paper synthesizes quantitative findings from multiple studies (primarily Steinicke et al. and Jerald et al.) regarding how much manipulation is "safe" (imperceptible).

• Rotation Gains:

  • Users are less sensitive when the gain is in the direction of the head turn.
  • Key Finding: Users can be physically turned approximately 49% more and 20% less than the virtual rotation without noticing.
  • Implication: You can make a $90^\circ$ physical turn look like a $134^\circ$ virtual turn.

• Translation Gains:

  • Key Finding: Imperceptible scaling ranges from downscaling by 14% to upscaling by 26%.
  • Implication: A 10m virtual hallway can be walked in a physical space of roughly 8m to 11.6m.

• Curvature Gains:

  • Key Finding: Results vary wildly based on walking speed and methodology.
  • Radius: Estimates for the radius of a circle users can be steered onto (while thinking they are walking straight) range from 22 meters (conservative) down to 6.4 meters (aggressive).
  • Implication: To allow truly infinite straight-line walking in VR, you need a physical lab space of at least $15m \times 15m$ to $45m \times 45m$ to ensure the circles fit. This is too big for most homes.

• Bending Gains:

  • Key Finding: A virtual curve can be bent up to 4.4 times its radius in the real world.

6.2. Side Effects Analysis

6.2.1. Simulator Sickness

• Findings: Studies generally show an increase in simulator sickness scores (SSQ) when RDW is used compared to baseline, but often it is not severe.
• Nuance: It is difficult to separate sickness caused by RDW from sickness caused by VR hardware latency.
• Audio-Visual: Adding spatial sound might actually increase sickness if the audio cues conflict with the visual/vestibular manipulation.

6.2.2. Spatial Memory

• Positive Result: Hodgson et al. found no negative impact of RDW on spatial memory (landmark learning).
• Comparison: Users using RDW (even with slight redirection) performed better on spatial tasks than those using joysticks or walking-in-place. This confirms that the act of physical walking is crucial for spatial cognition, even if that walking is manipulated.
• Negative Result (Overt): Overt interventions (like the 2:1 turn) do harm spatial memory because they break the flow and disorient the user.

6.2.3. Cognitive Load

• Findings: Bruder et al. found that RDW does increase cognitive load, particularly Curvature Gains.
• Relationship: The tighter the circle (smaller radius), the higher the cognitive load. This suggests the brain is actively working to resolve the sensory conflict, drawing resources away from working memory.

6.3. Architectural Manipulation Results

• Change Blindness: Highly effective. In one study, only 1 out of 77 participants noticed that a door had moved.
• Impossible Spaces: Rooms can overlap by 56% (for small rooms) or 31% (for large rooms) without detection. This allows for significant compression of indoor environments.

7. Conclusion & Reflections

7.1. Conclusion Summary

The paper concludes that Redirected Walking (RDW) is a mature but evolving field. It has successfully proven that:

1. Imperceptible manipulation is possible: We can decouple virtual and physical motion significantly without users knowing.
2. Real walking is superior: Even with manipulation, RDW preserves the benefits of real walking (presence, spatial memory) better than artificial locomotion techniques.
3. Controllers are key: The challenge has shifted from "can we do it?" to "how do we automate it?" using predictive and generalized controllers.

7.2. Limitations & Future Work

The authors identify several critical open challenges:

• Space Requirements: Current thresholds (e.g., 22m radius for curvature) are still too large for typical consumer "room-scale" VR ($3m \times 3m$).
• Generalizability: Most controllers require pre-processing or specific environments. There is a need for "plug-and-play" controllers that work in any VE.
• Multi-User Redirection: Redirecting two people in the same physical room without them colliding is a massive computational and safety challenge.
• Individual Differences: We don't know enough about how sensitivity varies across the population (age, gamer experience) to customize gains dynamically.
• Safety in "Building-Scale": As VR moves to larger, cluttered physical spaces (warehouses, offices), RDW needs to account for dynamic physical obstacles, not just empty labs.

7.3. Personal Insights & Critique

• The "Home User" Gap: While RDW is fascinating, the paper implicitly reveals a hard barrier for consumer adoption: the physics of human perception. If we need a 20m radius to trick the vestibular system, software alone cannot solve the "VR in a small apartment" problem purely with curvature gains. This suggests that Architectural Manipulation (Impossible Spaces) or Overt Interventions (Distractors) might be the only viable path for consumer VR, rather than the "pure" subtle continuous redirection.
• Cognitive Load Trade-off: The finding that RDW increases cognitive load is critical. For training simulations (e.g., firefighters), this is fine. But for entertainment or education, if the locomotion technique makes the user "dumber" (reduces working memory), it might interfere with the primary goal of the application.
• Potential Application: The "Masking" techniques (using blinks/saccades) are under-utilized. With modern eye-tracking in headsets (like the PSVR 2 or Apple Vision Pro), these techniques could become standard, applying tiny corrections constantly without the user ever knowing, potentially lowering the required space threshold.