Michael J. Tarr: Exploring Visual Navigation and Object Recognition with Virtual Reality: Early Lessons from the VENLab

We have developed a large-scale (50'x50') virtual environment (the VENLab) for studying questions in visual navigation and object recognition. Unique to our facility is the ability of subjects to actively walk around an artificial environment in which we can carefully control the position and the appearance of objects. The first experiments we have run in the VENLab address four questions: 1) What is the role of visual flow information in different regions of the scene with regard to the accuracy of dead reckoning ? 2) What are the relative roles of dead reckoning and landmarks in navigation? And are local and global landmarks weighted equally? 3) Are there differences in the spatial updating that occurs with regard to viewpoint when orientation changes are generated by observer motion as opposed to object rotation? 4) Does falling off a 50' virtual cliff hurt?