From: John Wann Subject: SCI: Visual Cues to Depth Perception In reply to Mike (Koopman's) point- Mike asks: > However, doesn't "... consideration of what features are essential >to support human control in the task to be undertaken," require >an understanding of the cognitive processes through which this >operator feedback control (telepresence) is implemented. > Yes, although "cognitive processes" is a term currently out of favour with many perception researchers. "perceptual processes" certainly. > > Doesn't the quest for "reality" further an understanding of the >impact of sensory data which is "... essential to support human > control in the task to be undertaken." > I'm not convinced of this, the more features of the real visual scene you bundle in then the more "real" it will look, but this doesn't tell you what you can leave out. If you can get away without stereo for instance, you can make a great computational saving, but possibly reduce the reality content. I don't think bundling in features and using the "that looks great" metric will give you big gains in the control vs complexity problem, although "looking good" is what currently sells systems. What you need is to define the task for a given VE system; e.g. The precise placement and manipulation of small objects for micro-surgery Then decide on what are critical measures of (natural) peformance; e.g. terminal accuracy, control of velocity, control of grasp force. Then monitor these measures as different features are included/excluded or perturbed (introducing a perturbation can give you an indication of the extent to which the operator is relying on that info source). The problem with reality is there is too much information and to try and duplicate this content within a VR is not the most sensible route to building an effective system for an applied task. It maybe of course that a developer doesn't have an "applied" goal, but then we come back to the question of what do we want VR for ? John Wann Edinburgh Visual Rhetoric Project