Optimal Planning for Redirected Walking Based on Reinforcement Learning in Multi-user Environment with Irregularly Shaped Physical Space

Abstract

Fig. 1: Reset control during multi-user RDW experience in a physical space with a circle obstacle in the center: (a) Reset-to-Center (R2C): the user unconditionally reset to the center of the physical space without considering other users or obstacle in the center. (b) Reset-to-Gradient (R2G): the user reset to the direction of the sum of the force that is acted on them by the boundary of the physical space, obstacles, and other users. (c) Multi-user Reset Controller (MRC): the user reset to the optimal direction determined by a model trained with information about the user's physical space.

Effects of Virtual Room Size and Objects on Relative Translation Gain Thresholds in Redirected Walking

“…Dong et al [16] introduced Smooth Assembly Mapping (SAM), which decomposes large VE to smaller local patches and mapping together into a real workspace. For multi-user scenarios, Azmandian et al [4] explored immersive VR experiences for redirected walking for two users in the same physically tracked space, and Lee et al [27] used reinforcement learning in a multi-user environment with heterogeneous physical space.…”

ENI: Quantifying Environment Compatibility for Natural Walking in Virtual Reality

“…Moreover, non-convex PEs like the cross and L-shaped rooms lead to more collisions than the convex PEs. Lee et al [22] also studied how RDW algorithms perform as the shape and size of the PE changes. In their work, they considered square PEs of varied sizes, as well as PEs in the shape of a square, trapezoid, cross, circle, T, and L, each with a roughly equal area.…”

“…In their work, they considered square PEs of varied sizes, as well as PEs in the shape of a square, trapezoid, cross, circle, T, and L, each with a roughly equal area. Lee et al [22] observed that larger PEs lead to fewer collisions and that non-convex PE shapes like cross, T, and L lead to more collisions than the convex PEs. Williams et al [43] introduced the Complexity Ratio (CR) metric to quantify the navigability of a PE, VE pair.…”

“…Redirection controllers are algorithms that are used to compute the correct redirection that is applied to optimally steer the user away from physical objects [24]. The performance of controllers is typically evaluated by measuring the number of times the user is reset after getting too close to a physical object [4,23,38,43,44], the average distance the user is able to walk before initiating a reset [4,35,38,43], or the average strength of redirection during locomotion [4,22,43]. These performance metrics are useful, but they do not provide any information in terms of how similar is the user's proximity to obstacles in the PE and VE.…”

Understanding, Modeling and Simulating Unintended Positional Drift during Repetitive Steering Navigation Tasks in Virtual Reality

“…For instance, simulations have been performed to compare different redirection strategies [19], design new redirection methods [5,43], optimize obstacle avoidance in constrained workspaces [42,48], study the size and shape of the workspace [3,24], or minimize collisions between several redirected users in the same workspace [4,5,23]. Where most implementations of redirection techniques rely on human-engineered logic, it is worth noting that recent work showed that machine-learning with simulation approaches could outperform the state of the art steering algorithms used for redirected walking [23,36,41].…”