Intelligent Pre-Reset for Redirected Walking through Reinforcement Learning

Redirected walking (RDW) algorithms subtly manipulates users’ movements in virtual environments through three gains, distorting their trajectories in the physical space, thus allowing them to explore infinite virtual environments within limited physical spaces. However, collisions with physical boundaries or obstacles in physical spaces are inevitable. At this point, explicit redirected techniques, such as reset, are needed to pause users’ roaming experience and redirect them to open areas. Currently, most reset strategies wait until a collision occurs before resetting the user, while these reset algorithms can achieve certain effectiveness in simple physical environments. However, in complex physical environments, users may walk into narrow areas before collisions occur. Therefore, no matter which direction they turn, further collisions are likely to happen over short distances. To enhance obstacle avoidance capability while adapting to various complex physical layouts, this paper adopts the concept of pre-reset and presents a novel method of intelligent pre-reset IPR for redirected walking through reinforcement learning, which determines whether an early reset should be performed and the optimal reset direction based on the current physical environment, as well as the user’s position and walking trajectory. Through simulation and human experiments compared with other reset algorithms in several environments, we demonstrate the superiority of our approach, achieving higher average walking distances between resets while adapting to various nuances of reset techniques. At the same time, IPR is applicable to various RDW algorithms and environments of different complexities.