Register now After registration you will be able to apply for this opportunity online.
Predictive Resets using Path Prediction
Walking has been proven to create the best sense of presence while exploring
virtual environments. However, walking in virtual environments comes with the
constraint that the virtual and the physical space must have the same
dimensions. This limitation restricts the size of the virtual spaces. Various techniques have been developed to overcome this issue, such as RDW and impossible spaces. Although significant progress has been made with these methods, exploring large virtual environments can still lead the user to reach the boundaries of the physical space. Therefore, a safety mechanism is needed to prevent the user from colliding with the walls. The most popular safety mechanism is having resets, which are messages displayed to the user asking them to stop and perform a certain action that turns them away from the wall before continuing the exploration. However, resets reduce immersion and thus should be avoided as much as possible.
In this project, you will extend a prior study, where evaluated the application of reinforcement learning for predicting the need for new resets and their respective directions. While it didn't outperform state-of-theart methods in reducing the number of resets, the potential of the technology was evident, particularly when testing various reward functions, which significantly influenced performance. In this study, we aim to refine our previously implemented algorithm by adding path prediction to anticipate the need for an early reset. You will test and evaluate your algorithm across a series of simulations. You will present your work in an intermediate and a final presentation to the ICVR lab. Finally, you will summarize your results in a written report.
In this project, you will extend a prior study, where evaluated the application of reinforcement learning for predicting the need for new resets and their respective directions. While it didn't outperform state-of-theart methods in reducing the number of resets, the potential of the technology was evident, particularly when testing various reward functions, which significantly influenced performance. In this study, we aim to refine our previously implemented algorithm by adding path prediction to anticipate the need for an early reset. You will test and evaluate your algorithm across a series of simulations. You will present your work in an intermediate and a final presentation to the ICVR lab. Finally, you will summarize your results in a written report.
• Literature research on resets and path prediction
• Familiarize yourself with the previously developed code
• Enhance the existing algorithm by integrating user path prediction
• Conduct a user study where your algorithm is compared to other sota reset techniques
• Intermediate and final presentation
• Written report
• Literature research on resets and path prediction • Familiarize yourself with the previously developed code • Enhance the existing algorithm by integrating user path prediction • Conduct a user study where your algorithm is compared to other sota reset techniques • Intermediate and final presentation • Written report
Mathieu Lutfallah, CLT D14 lutfallah@iwf.mavt.ethz.ch
Andreas Kunz, CLT C13 kunz@iwf.mavt.ethz.ch
Mathieu Lutfallah, CLT D14 lutfallah@iwf.mavt.ethz.ch Andreas Kunz, CLT C13 kunz@iwf.mavt.ethz.ch