Long-short term memory prediction of user’s locomotion in Virtual Reality

Virtual reality has some well-known limitations when the user walks naturally. A user experiencing virtual reality may collide with the limits of the physical workspace. To solve this, there exist techniques, such as the redirected walking methods, that try to trick the human brain to avoid these limits. However, these methods require good predictions of where users want to go in the future. On the other hand, Deep learning is helping in so many areas to make accurate predictions. Specifically, Long short-term memory obtained promising results in this virtual reality prediction task giving us clues to continue researching in this line. This manuscript presents an experiment carried out with 44 participants collecting positional data. These data have been used to test different existing prediction algorithms. In addition, a way to improve the previous results obtained with deep learning is presented. These results are achieved with the use of rotation quaternions and various positional data in three dimensions. The authors strongly believe that there is still much room for improvement in this research area.