Performance of visual tracking systems: implications for visual controlled motion

In visual control of motion the performance of the system is typically tied to the speed at which the visual information can be processed. Visual processing is unusual in that the data is not the information and in some sense the quality of the information can often be improved through extended processing (e.g. the resolution). A visual sensor tracking an object will only provide sporadic observations due to the processing delay, and, computational delay can pose difficulty within the control structure. We explore the trade off between the computational delay, the effective system observability, the accuracy of the visual processing, and implications for visual-guided motions. The task of visually tracking a target in order to localize it with sufficient accuracy to "touch" it provides a demonstration of the trade-off between computational delay and the motion (visual localization) accuracy.