Machine Learning Applied to Locomotion in Virtual Reality

The objective of this research project is to recognize a user’s walking pattern on a treadmill-like platform for navigation in Virtual Reality (VR) environments. To achieve this, a walking recognition software solution was developed using Convolutional Neural Networks (CNNs), a type of Machine Learning (ML) architecture. The ML model was trained on a dataset collected from users’ movements in a virtual simulation, tracking the positions and rotations of their feet as they walked in various directions and orientations. Tracking was accomplished using 6 degrees of freedom (6DoF) trackers placed on the users’ feet. The neural network achieved a 94% accuracy rate during testing. Due to the network’s lightweight configuration, the response time is, on average, 0.1 seconds, demonstrating its potential as an efficient natural controller for real-time user navigation in multiple directions.