Automatic Path Generation for Assisted Navigation in Virtual Environments

  • By @shailen@
  • 25 September, 2013
  • Comments Off on Automatic Path Generation for Assisted Navigation in Virtual Environments

3D interfaces are hard to navigate using conventional techniques for new users. Hence using assisted navigation paths learnt from expert motions or from a set of large number of moves can greatly improve a user experience in a Virtual Environment.

The first problem is reconstruction of navigational paths from just tile hit data, which stores the user ip and time of hit for a particular tile ( A tile is one of the images stored on a server to display information regarding maps and images in a heriarchichal manner )

Once we have the paths the next step is to learn good paths from the sets of paths we have. Also most of this paths pass through same regions which are classified as regions of high interest. Thus when a new user now navigates the same virtual environment, these paths can be used to tweak the moves performed by the user. Also the regions of high interest can be pointed out to turn the user’s attention.

Below are presented some results using Microsoft’s Virtual Earth 3D engine and data logs of tile visits by users.

These images depict the paths regenerated from user logs which are then used to determine paths which will be used in assisted navigation

The use of these techniques has also been shown to be of significant impact on training of bots using multiplayer moves by Soumyajit Deb, Microsoft Research. Using a database of moves performd by expert players the bots can be trained using just a new AI training routine. The huge database of expert moves already present can thus be used to improve results of bot movement and intelligence.

Q3A Tracklog

Due to the ongoing nature of the project at Microsoft Research it is not possible to release more information at the moment.

Categories: Projects

Comments are closed.