Experiments were performed to observe the robot hand. The low level visual feature acquisition is performed on the Datacube MaxVideo pipelined video processor at frame rate. The observer and manipulating robots are both PUMA 560's and the Lord experimental gripper is used as the manipulating hand.
The experiment was shot with three video camera. The right hand side of the images show the actual observer and manipulation workspace and the different configurations as the experiment proceed. The upper left corner shows the observer view, which is the set of images grabbed by the camera for processing. The lower left corner shows the observer state, that is, what the observer ``thinks''. A graphical representation of the different states and their change is used. Fail states are represented by an empty box. Figures 13 and 14 illustrate a manipulation experiment. In this sequence the hand tries to insert a peg in a hole. The observer approaches and focuses on the peg and hole when the peg gets nearer to the hole. State changes occur when the hole appears and when insertion is asserted.